accurate modeling techniques for power delivery
Post on 15-Nov-2021
3 Views
Preview:
TRANSCRIPT
Scholars' Mine Scholars' Mine
Doctoral Dissertations Student Theses and Dissertations
Summer 2020
Accurate modeling techniques for power delivery Accurate modeling techniques for power delivery
Jingdong Sun
Follow this and additional works at: https://scholarsmine.mst.edu/doctoral_dissertations
Part of the Electrical and Computer Engineering Commons
Department: Electrical and Computer Engineering Department: Electrical and Computer Engineering
Recommended Citation Recommended Citation Sun, Jingdong, "Accurate modeling techniques for power delivery" (2020). Doctoral Dissertations. 2921. https://scholarsmine.mst.edu/doctoral_dissertations/2921
This thesis is brought to you by Scholars' Mine, a service of the Missouri S&T Library and Learning Resources. This work is protected by U. S. Copyright Law. Unauthorized use including reproduction for redistribution requires the permission of the copyright holder. For more information, please contact scholarsmine@mst.edu.
ACCURATE MODELING TECHNIQUES FOR POWER DELIVERY
by
JINGDONG SUN
A DISSERTATION
Presented to the Graduate Faculty of the
MISSOURI UNIVERSITY OF SCIENCE AND TECHNOLOGY
In Partial Fulfillment of the Requirements for the Degree
DOCTOR OF PHILOSOPHY
in
ELECTRICAL ENGINEERING
2020
Approved by:
Jun Fan, Advisor DongHyun Kim
Victor Khilkevich Chulsoon Hwang Yaojiang Zhang
© 2020
Jingdong Sun
All Rights Reserved
PUBLICATION DISSERTATION OPTION
iii
This dissertation consists of the following three articles, formatted in the style used
by the Missouri University of Science and Technology:
Paper I, found on pages 7-39, is conditionally accepted with major revisions in
IEEE Transactions on Electromagnetic Compatibility.
Paper II, found on pages 40-72, has been submitted to IEEE Transactions on Power
Electronics.
Paper III, found on pages 73-116, has been published in IEEE Transactions on
Power Electronics, vol. 35, no. 8, pp. 7840-7852, Aug. 2020.
iv
ABSTRACT
Power delivery is essential in electronic systems to provide reliable power from
voltage sources to load devices. Driven by the ambitious user demands and technology
evolutions, the power delivery design is posed serious challenges. In this work, we focus
on modeling two types of power delivery paths: the power distribution network (PDN) and
the wireless power transfer (WPT) system.
For the modeling of PDN, a novel pattern-based analytical method is proposed for
PCB-level PDN impedance calculations, which constructs an equivalent circuit with one-
to-one correspondences to the PCB’s physical structure. A practical modeling
methodology is also introduced to optimize the PDN design. In addition, a topology-based
behavior model is developed for the current-mode voltage regulator module (VRM). This
model includes all the critical components in the power stage, the voltage control loop, and
the current control loop of a VRM device. A novel method is also proposed to unify the
modeling of the continuous and discontinuous conduction modes for transient load
responses. Cascading the proposed VRM model with the PCB-level PDN model enables a
combined PDN analysis, which is much needed for modern PDN designs.
For the modeling of WPT system, a system-level model is developed for both
efficiency and power loss of all the blocks in WPT systems. A rectifier characterization
method is also proposed to obtain the accurate load impedance. This model is capable of
deriving the power capabilities for both the fundamental and higher order harmonics. Based
on the system model, a practical design methodology is introduced to simultaneously
optimize multiple system parameters, which greatly accelerates the design process.
v
ACKNOWLEDGMENTS
Throughout my doctoral studies and writing of this dissertation, I have received a
great deal of support and assistance from my mentors, collaborators, family, and friends.
I would like to express my deepest gratitude to my advisor, Dr. Jun Fan, for his
guidance, encouragement, and trust of my study. Dr. Fan not only gave me insightful
comments on my research projects, but also helped me develop growth mindsets and
positive attitudes towards my future career.
I would like to thank Dr. DongHyun Kim, Dr. Victor Khilkevich, Dr. Chulsoon
Hwang, and Dr. Yaojiang Zhang for serving on my doctoral committee and providing
helpful feedbacks on my research work.
In addition, I would like to extend my appreciation to all the faculty members and
staffs in the UMR/MST EMC Lab. I would like to specially thank Dr. James Drewniak for
inspiring my career path. My sincere thanks also go to Dr. Jonghyun Cho, Dr. Heegon Kim,
and Dr. Hongseok Kim for their advice and assistance in my projects. I truly appreciate the
help and coorperation from the students in the EMC lab, in particular Anfeng Huang, Xin
Yan, Ze Sun, Shengxuan Xia, and Nicholas Erickson.
Finally, I am deeply grateful to my parents, Fengwu Sun and Xiuhong Wang for
their constant support. I would like to express my love for my girlfriend, Wei Li, who
provides me continuous encouragements in the past six years. This dissertation would not
have been possible without their love and endless support.
This dissertation is based upon work supported partially by the National Science
Foundation (NSF) under Grant No. IIP-1916535.
vi
TABLE OF CONTENTS
Page
PUBLICATION DISSERTATION OPTION.......................................................................... iii
ABSTRACT.................................................................................................................................. iv
ACKNOWLEDGMENTS........................................................................................................... v
LIST OF ILLUSTRATIONS.......................................................................................................x
LIST OF TABLES.....................................................................................................................xiii
SECTION
1. INTRODUCTION..............................................................................................................1
1.1. POWER DISTRIBUTION NETW ORK..................................................................2
1.2. WIRELESS POWER TRANSFER SYSTEM......................................................... 3
1.3. CONTENTS AND CONTRIBUTIONS...................................................................5
PAPER
I. A PATTERN-BASED ANALYTICAL METHOD FOR IMPEDANCE CALCULATION OF THE POWER DISTRIBUTION NETWORK IN MOBILE PLATFORM S....................................................................................................................... 7
ABSTRACT......................................................................................................................... 7
1. INTRODUCTION.......................................................................................................... 8
2. PATTERN FORMULATION AND CALCULATION.......................................... 12
2.1. PATTERN FORMULATION............................................................................. 12
2.2. PATTERN CALCULATION.............................................................................. 16
2.3. SIMULATION STUDIES ON ACCURACY AND EDGE-EFFECT.........22
3. PDN MODELING METHODOLOGY 24
3.1. PCB DIVISION.....................................................................................................27
3.2. PATTERN IDENTIFICATION......................................................................... 28
3.3. CIRCUIT RECONSTRUCTION....................................................................... 29
3.4. DESIGN OPTIMIZATION................................................................................. 31
4. VALIDATION.............................................................................................................. 31
4.1. WHOLE STRUCTURE COMPARISON......................................................... 33
4.2. REGION-BY-REGION-COMPARISON......................................................... 35
5. CONCLUSION............................................................................................................. 36
REFERENCES ................................................................................................................ 37
II. TOPOLOGY-BASED ACCURATE MODELING OF CURRENT-MODE VOLTAGE REGULATOR MODULES FOR POWER DISTRIBUTION NETWORK DESIGN.......................................................................................................40
ABSTRACT....................................................................................................................... 40
1. INTRODUCTION........................................................................................................41
2. VRM TOPOLOGY.......................................................................................................44
3. TOPOLOGY-BASED GENERIC BEHAVIOR M ODEL..................................... 47
3.1. VOLTAGE CONTROL LOOP.......................................................................... 47
3.2. CURRENT CONTROL LOOP...........................................................................50
3.3. POWER STAGE...................................................................................................53
3.4. MODEL IMPLEMENTATION......................................................................... 54
4. CHARACTERIZATION AND VALIDATION.......................................................56
4.1. MEASUREMENT SETUP................................................................................. 57
4.2. PARAMETER OPTIMIZATION...................................................................... 58
vii
4.3. MODEL VALIDATION 62
5. CONCLUSION........................................................................................................... 68
APPENDIX........................................................................................................................ 68
REFERENCES..................................................................................................................70
III. ACCURATE RECTIFIER CHARACTERIZATION AND IMPROVED MODELING OF CONSTANT POWER LOAD WIRELESS POWER TRANSFER SYSTEMS................................................................................................. 73
ABSTRACT....................................................................................................................... 73
1. INTRODUCTION........................................................................................................74
2. TOPOLOGY AND PREINVESTIGATION............................................................ 77
2.1. INVESTIGATION OF SYSTEM AND COIL EFFICIENCY...................... 81
2.2. INVESTIGATION OF HARMONIC FREQUENCY.................................... 81
2.3. INVESTIGATION OF RECTIFIER IM PEDANCE.......................................83
3. ACCURATE RECTIFIER CHARACTERIZATION............................................. 84
4. IMPROVED MODELING AND DESIGN METHODOLOGY............................92
4.1. IMPROVED MODELING OF THE WPT SYSTEM..................................... 93
4.2. PRACTICAL DESIGN METHODOLOGY...................................................100
5. EXPERIMENT AND VALIDATION..................................................................... 102
5.1. VALIDATION ON CHARACTERIZATION AND MODELING...........104
5.2. VALIDATION ON DESIGN METHODOLOGY.........................................108
6. CONCLUSION........................................................................................................... 110
APPENDIX...................................................................................................................... 110
viii
REFERENCES 113
ix
SECTION
2. CONCLUSIONS............................................................................................................117
BIBLIOGRAPHY......................................................................................................................119
VITA........................................................................................................................................... 122
LIST OF ILLUSTRATIONS
SECTION Page
Figure 1.1. Power delivery paths in electronic devices...........................................................1
PAPER I
Figure 1. Comparison of PCB in mobile platforms and PCB in other platforms................. 9
Figure 2. Major improvements of the proposed pattern-based analytical method............12
Figure 3. Localized via pattern around the power via.............................................................13
Figure 4. Formulated via patterns of two categories...............................................................14
Figure 5. Magnetic flux though the loop formed by adjacent vias....................................... 16
Figure 6. Illustration of the magnetic flux for an arbitrary via pattern.................................18
Figure 7. Arbitrary via patterns for validation....................................................................... 21
Figure 8. Comparison of via inductance for different patterns and distances.....................23
Figure 9. Simulation results of via inductance for different via lengths..............................24
Figure 10. Flow diagram of the modeling methodology........................................................25
Figure 11. Physical geometry and region division of the mobile phone PCB.................... 26
Figure 12. The equivalent circuit corresponding to the PCB under investigation............ 27
Figure 13. Selection of the optimal pattern size......................................................................29
Figure 14. Calculation of the equivalent inductance and resistance in one region........... 30
Figure 15. Photograph and diagram of the shunt-thru measurement setup......................... 32
Figure 16. Comparison of the PDN impedance for the whole PCB structure.....................34
PAPER II
x
Figure 1. Main elements in the end-to-end PDN 42
xi
Figure 2. System-level PDN performance............................................................................... 43
Figure 3. Diagram of an open-loop dc-dc buck converter..................................................... 45
Figure 4. Typical feedback control topologies in VRM.........................................................45
Figure 5. Modeling of the voltage control loop.......................................................................49
Figure 6. The inductor current in one switching period......................................................... 51
Figure 7. Schematic of the behavior model in the circuit simulator.....................................55
Figure 8. Diagram and photograph of the measurement setup..............................................58
Figure 9. Measured voltage and current waveforms for characterization............................59
Figure 10. Validation of the unified equations for the CCM and DCM.............................. 63
Figure 11. Comparison of voltage droop waveforms for single-phase VRM.....................65
Figure 12. Comparison of voltage droop waveforms for 3-phase VRM............................. 66
Figure 13. Comparison of simulated and measured impedance for single-phase VRM. . 67
PAPER III
Figure 1. Typical WPT system with the SS topology...........................................................74
Figure 2. Diagram of the circuit model for preinvestigation................................................78
Figure 3. Simulation results of the circuit model...................................................................80
Figure 4. Circuit model of a full-wave diode rectifier.......................................................... 85
Figure 5. Comparison of simulated and measured rectifier impedance............................. 86
Figure 6. Measurement-based rectifier impedance characterization setup........................ 89
Figure 7. Rectifier impedance characterization and validation results...............................91
Figure 8. Formulation for the improved modeling of the WPT system............................. 94
Figure 9. Time-domain voltage and current waveforms at the switching node................ 97
Figure 10. Flowchart of the design methodology................................................................101
Figure 11. Photograph of the experimental prototype........................................................ 103
Figure 12. Comparison of dc power and efficiency............................................................ 105
Figure 13. Comparison of intermediate ac powers..............................................................107
Figure 14. FOM calculation in the initial phase.................................................................. 109
Figure 15. System efficiency comparison in the fine-tune phase......................................109
xii
LIST OF TABLES
xiii
PAPER I Page
Table 1. Region-by-region comparison of parasitic elements..........................................35
PAPER II
Table 1. VRM behavior model parameters............................................................................ 56
Table 2. Characterized single-phase and 3-phase VRM models......................................... 61
PAPER III
Table 1. Coil parameters for characterization and validation setups.................................. 89
Table 2. Categories of the WPT system model parameters................................................. 93
1. INTRODUCTION
Power delivery is essential in electronic systems to provide reliable power from
voltage sources to load devices. Driven by the ambitious user demands and technology
evolutions, the operation voltage of active devices is progressively lower to meet faster and
more efficient data processing requirements [1]. In addition, modern integrated circuits
(ICs) are dealing with more sophisticated scenarios. As a result, the amplitude and slew
rate of load currents keep increasing while the voltage tolerance margin becomes tighter
[2]. Therefore, the power delivery design is posed new challenges in efficiency, stability,
and impedance optimizations [3-5].
DecouplingCapacitor PCB
Planes - Vias — Voltage Regnlatoi
---------- Module
Package and Die
Batten’Power Distribution > etwork
Dc Voltage Source Inductoi----TfP
RX Coil Rectitiei Charger
MatchingMatchingNetworkN etwork
PoweiAmplifier IX Coil
wireless PowerTransfer System
Figure 1.1. Power delivery paths in electronic devices.
2
As depicted in Figure 1.1, the power delivery paths consist of various components,
such as power converters, inductors, decoupling capacitors, power-ground planes in
printed circuit boards (PCBs), and vias between PCB layers. It is desired to accurately
model all the key components and propose pratical optimization methodologies. In this
dissertation, we focus on modeling two common types of power delivery paths: the power
distribution network (PDN) and the wireless power transfer (WPT) system. This section
introduces the background of the study, the state-of-the-art modeling methods, and the
major contributions of the included articles.
1.1. POWER DISTRIBUTION NETWORK
The primary objective of the PDN design is to ensure normal IC operations by
minimizing voltage fluctuations. As shown in the blue dot-dashed block of Figure 1.1, the
end-to-end PDN has four main parts: (1) voltage regulator module (VRM); (2) PCB-level
PDN; (3) package-level PDN; and (4) on-die PDN. The high frequency response is mainly
determined by the package/die parasitics (usually above 100 MHz), the middle frequency
response is affected by the PCB design (usually 300 kHz ~ 100 MHz), and the low
frequency response is typically dominated by the VRM (usually below 300 kHz). With the
increasingly complicated load current profile, a successful PDN design needs to optimize
all the parts and satisfy the target impedance from dc to high frequencies.
In the past few decades, the modeling of PDN has been extensively studied. For the
VRM modeling, the simple first-order linear resistor-inductor (RL) or the second order 4-
element RL circuits were commonly used to fit the output impedance [6]. However, these
discrete passive elements are not able to represent the control loop behaviors of modern
3
VRMs. A behavior model was proposed by simplying the control loop as a boost converter
[7]. This unrealistic simplication has no physical meaning and leads to inaccurate results.
A losed-loop mathematical model was developed to calculate the output impedance of
VRMs [8], but this approach is purely mathematical and not intuitive to provide insights to
improve the VRM design. The VRM vendors could provide encrypted device models using
the commercial circuit simulators, such as SIMPLIS [9]. However, this model is only
applicable to a specific device and the detailed implementations are confidential. For the
PCB-level and package-level PDNs modeling, the fast transmission-line method (TLM)
[10] and cavity model method [11-13] were commonly used to model the multilayered
PCB. The partial-element equivalent circuit (PEEC) could also be applied to the chip-
package hierarchical PDN structures [14-16]. In addition, full-wave electromagnetic
modeling methods, including the finite-difference time-domain (FDTD) [17], the finite-
element method (FEM) [18], and the method-of-moments (MOM) [19], were integrated
into simulations tools for PCB and package PDN impedance. However, the existing
methods may either become impractical or have limitations to model the PDN impedance
in some new applications, such as the PCB design in the mobile platform.
1.2. WIRELESS POWER TRANSFER SYSTEM
WPT is an emerging technology that brings great convenience to charging a large
variety of electronic devices. As shown in the red dashed block of Figure 1.1, a typical
WPT system consists of four parts: (1) the power amplifier which converts dc voltage to
ac power; (2) the transmitter (TX) coil with its matching network; (3) the receiver (RX)
coil with its matching network; and (4) the rectifier which inversely regulates ac power to
4
dc output voltage. There are also industrial standards for WPT applications to improve the
safety and interoperability, such as Wireless Power Consortium (WPC) Qi for consumer
wireless charging devices and Society of Automotive Engineers (SAE) J2954 for wireless
charging vehicles. In addition to the standards and regulations, it is crucial to minimize the
power loss and improve the efficiency of WPT systems.
Various studies have been carried out to investigate WPT designs at both coil and
system levels. A magnetic coil design method was presented in [20]. Other system
parameters, such as the input voltage level, the coil matching networks, the operating
frequency, and the load impedance, are also critical for efficiencies. The power transfer
efficiency and capability were analyzed under different matching networks in [21] and [22].
However, the output load of the coils was assumed to be purely resistive and the nonlinear
effects of the rectifier was not considered. The impacts of rectifier impedance were
discussed in [23] and [24], but the authors only focused on the resonant condition and
ignored the nonlinearity caused by the line-commutation process. Analytical expressions
of coil-to-coil efficiency for both frequency and voltage tuning WPT systems were
developed in [25]. However, the derivations were based on the First Harmonic
Approximation (FHA), which becomes inaccurate when the system is operating at off-
resonant frequency or the load quality factor is decreasing. In sum, it is desired to overcome
the impractical assumptions of existing methods and develop an accurate system-level
model for WPT applications.
5
1.3. CONTENTS AND CONTRIBUTIONS
This dissertation presents accurate modeling techniques for PDN and WPT systems.
The outlines and contributions of this dissertation are summarized.
In the first paper, a novel pattern-based analytical method for PDN impedance
calculation is proposed. This method focuses on the localized patterns formulated by
adjacent vias. Therefore, it can easily decompose the complicated PCB structures in mobile
platforms and avoid the extensive workload to identify the absolute positions of all the vias
and decoupling capacitors at different layers. In addition, the proposed method uses
analytical equations to calculate the impedance based on the localized via patterns, which
provides more flexibility to optimize the PDN design compared to the complicated
numerical solutions, especially in the pre-design stage. A practical modeling methodology
is also developed based on the pattern-based analytical model to optimize the PCB-level
PDN design. It constructs an equivalent circuit with one-to-one correspondences to the
PCB’s physical geometry, which is helpful to identify the most critical component
contributing to the total impedance and improve the overall PDN design.
In the second paper, a topology-based generic behavior model is developed for the
VRM in PDN design. This model includes all the interior topology-level components in
the power stage, the voltage control loop, and the current control loop of a VRM device.
Thus, it is helpful to provide more insights of the VRM’s operations to improve the PDN
design. A novel method is proposed to unify the modeling of the continuous and
discontinuous current modes (CCM and DCM) for different load conditions. It enables a
simplified procedure to implement the current control loop. In addition, a measurement-
based characterization method is proposed to optimize the model parameters. This method
6
can be applied to any provided VRM devices to bypass the confidential concerns. By
cascading the proposed VRM model with the PCB-level, the package-level, and the on-die
PDN models, a combined PDN analysis integrating the VR control can be established,
which is much needed for modern PDN designs.
In the third paper, a system-level model is developed for both dc-to-dc efficiency
and power loss of each interior part in WPT systems. This model is capable of deriving the
power capability for both the fundamental and higher order harmonics analytically, without
the impractical assumptions used by the existing methods. A rectifier characterization
method is also proposed to obtain the accurate load impedance for the system-level model.
This method is flexible and applicable to any provided rectifier devices. Based on the
proposed WPT model, a practical design methodology is introduced to simultaneously
optimize multiple system parameters, which greatly accelerates the design process
compared with the conventional trial-and-error method. They will become powerful tools
to improve the system performance and the thermal design of WPT applications.
7
PAPER
I. A PATTERN-BASED ANALYTICAL METHOD FOR IMPEDANCE CALCULATION OF THE POWER DISTRIBUTION NETWORK IN MOBILE
PLATFORMS
ABSTRACT
Power distribution network (PDN) is essential in electronic systems to provide
reliable power for load devices. With faster load transient current and lower voltage
tolerance margin for the microprocessors in mobile platforms, it is crucial to optimize the
PCB design to satisfy the strict target impedance. The conventional modeling methods
become impractical in mobile platforms due to the characteristics of high density
interconnect PCB and limited layout space. To overcome these issues, a pattern-based
analytical method for the PDN impedance calculation is presented. Based on the localized
patterns formulated by the relative relationships between the adjacent vias, the parasitic
elements are analytically determined for different regions of the entire PCB structure. With
the assistance of this method, a practical modeling methodology is developed to construct
an equivalent circuit with one-to-one correspondence to the PCB’s physical geometry. As
a result, the PDN design can be efficiently optimized especially for the pre-design stage to
accelerate the development process. Finally, the proposed method is validated through
measurements and full-wave simulations using a real mobile phone PCB in production.
8
1. INTRODUCTION
Power distribution network (PDN) is essential in electronic systems to provide
reliable power for load devices [1]. The primary objective of the PDN design is to ensure
normal IC operations by minimizing voltage fluctuations [2]. With the evolution in
technologies, modern active devices are being designed with lower voltage to meet faster
and more efficient data processing demands. As a result, the current draw and slew rate
keep increasing, while the voltage tolerance margin becomes tighter [3]. The mobile
platform, being a typical industrial application that follows these trends, is posed serious
challenges on the PDN design. Driven by the ambitious user requirements, more aggressive
architectures for mobile processors have been deployed to provide significant performance
improvements [4]. However, it also results in a fast transient current and a strict target
impedance at both dc and higher frequencies [5]. Furthermore, the high density
interconnect (HDI) PCB is commonly used in mobile platforms to increase the circuitry
density. The complicated interconnect structure brings non-negligible parasitic effects. In
addition, the slim form factor of mobile phones limits the layout space and the number of
surface mount decoupling capacitors, which causes difficulties to lower the PDN
impedance. Thus, the PCB-level PDN design, including the parasitic elements associated
with vias/power-ground planes and the decoupling capacitors, becomes more and more
critical for mobile platforms.
Over the past few decades, PDN modeling has been extensively investigated. First,
full-wave electromagnetic modeling methods were applied to study this problem, including
the finite-difference time-domain method (FDTD) [6], the finite-element method (FEM)
9
[7], the method-of-moments (MOM) [8], and the partial-element equivalent circuit (PEEC)
[9]-[11]. In addition, a fast and straightforward transmission-line method (TLM)
compatible for SPICE implementation was developed to model planes with bypass
capacitors [12]-[14]. Another fast method based on the cavity model was commonly used
to calculate the input impedance between the power-ground planes [15], [16]. The
reduction techniques for cavity model were also proposed to combine parasitic via
inductance and generate equivalent circuits [17], [18]. However, due to the characteristics
of the PCB in mobile phones, the existing methods either become impractical or have
limitations to model the PDN in mobile platforms.
PCB in Mobile Platforms PCB in Other Platforms
F a b r ic a t io n H ig h d e n s i ty in te rc o n n e c t P C B R e g u la r o r h ig h d e n s i ty in te rc o n n e c t P C B
P C B S h a p e I r re g u la r sh ap e R e c ta n g u la r sh ap e
V ia S iz e M ic ro -v ia , - 0 .0 7 5 m m d ia m e te r T h ro u g h -h o le v ia , - 0 .2 5 4 m m d ia m e te r
C a p a c i t o r s 2 0 - 3 0 , m o s tly in 0201 p a c k a g e 3 0 , m ix e d 0 6 0 3 /0 4 0 2 /0 2 0 1 p a c k a g e s
Figure 1. Comparison of PCB in mobile platforms and PCB in other platforms.
10
Figure 1 compares the PCBs in mobile platforms and other common platforms,
such as server or personal computer. The interconnect structure of PCB can be very
complicated, especially for the HDI PCB in mobile platforms. As shown in Figure 1 (a),
the 0.075 mm diameter micro-vias in mobile platform PCB could result in a huge number
of mesh cells for full-wave electromagnetic modeling methods. Hence, the computational
burden is heavy and the simulation time is long. Similarly, due to the complexity of the
geometry, the TLM model has to use a large number of circuit elements to converge with
good simulation accuracy. Figure 1 (b) illustrates the difference in via types. All the vias
in mobile platform PCB are blind/buried, while those in regular PCB are through-hole. The
cavity model method, which requires the positions of all the vias as the model’s ports, is
only suitable for the through-hole case. For through-hole vias, the port locations for all the
cavities formed by different plane pairs are exactly the same. Thus, the positions only need
to be identified once using one cavity, then they can be applied to other cavities for further
calculation. However, the blind/buried vias may be removed or shifted between cavities to
make room for routing traces within a limited layout space. In this case, the via positions
have to be determined separately for every cavity, which significantly increases the
workload. In addition, the cavity model method cannot handle the “dog-bone” traces
between the blind/buried vias easily. Another factor related to the port locations is depicted
in Figure 1 (c). Since the decoupling capacitors should be placed close to the power/ground
vias to minimize the parasitic inductance, the positions of decoupling capacitor’s
power/ground pads can be used to identify the port locations of the cavity model. For the
regular PCB with sufficient layout space, the decoupling capacitors are placed in order with
fixed pitch size and the same direction, so their positions can be easily determined. But for
11
the mobile platform PCB with limited layout space, the decoupling capacitors are
arbitrarily placed with random directions. It is difficult to find the absolute positions for all
the decoupling capacitors. Therefore, the cavity model method is impractical to model the
PCB in mobile platforms. At last, technical change and new product proliferation have
made the mobile phone industry extremely dynamic [19], which places greater demands
on accelerating the development process. In spite of the existing methods that only focus
on the post-validation stage (PCB layout is completed), the PDN design and modeling in
the pre-design stage are rarely investigated. Thus, it is desired to develop a PDN modeling
method that is suitable for mobile platforms and applicable in the pre-design stage.
To overcome the limitations of the existing methods for the PCB in mobile
platforms, a novel pattern-based analytical method is proposed for PDN impedance
calculation. Our method aims to achieve three major improvements, as shown in Figure 2.
Firstly, instead of using the absolute positions of all the vias and decoupling capacitors, the
parasitic elements are calculated based on the localized patterns. The formulation of pattern
and its benefits are discussed in Section 2. Secondly, the complicated numerical solutions
lack flexibility in the pre-design stage, so we integrate the analytical equations and
introduce a modeling methodology applicable to both pre-design and post-validation stages
in Section 3. Thirdly, in addition to modeling the whole PCB structure, our method is
capable of conducting region-by-region analysis. These regions still maintain one-to-one
correspondences to the physical geometry. So it is helpful to identify the dominant factors
in PDN design. In Section 4, our method is validated by both whole structure and region-
by-region comparisons using a real mobile phone PCB in production. Section 5 concludes
this article.
12
Figure 2. Major improvements of the proposed pattern-based analytical method.
2. PATTERN FORMULATION AND CALCULATION
A novel approach to calculate the parasitic elements of vias is presented in this
section based on the localized via patterns. Compared with the conventional methods using
the absolute via positions, our method possesses better flexibility to handle the complicated
PCB structures in mobile platforms. The accuracy of our method is validated through RF
simulations under various conditions. Since the via pattern is a 2-D structure, the sensitivity
of the edge-effect is also investigated.
2.1. PATTERN FORMULATION
As discussed earlier, the conventional PDN modeling methods become impractical
in mobile platforms. It is desired to develop a new modeling method based on the
characteristics of the mobile platform PCB. With the limited layout space and tight trace
routing restriction, the via-in-pad technique is extensively used for space optimizations. As
a result, the relative relationships (distance and angle) between the adjacent vias on the top
and bottom layers can be directly obtained from the IC’s pinout and decoupling capacitor’s
13
footprint. If the blind/buried via is shifted or removed on the internal layers, only the moved
via needs to establish new relative distance and angle with its adjacent vias, while the
relative relationships of all other vias remain the same. Therefore, the relative distances
and angles of the power and ground vias can be easily determined across all layers. Another
attribute of the mobile platform PCB, as shown in Figure 1 (a), is the usage of decoupling
capacitors with the same package. Restricted by the limited space, almost all the surface
mount decoupling capacitors use the small 0201 package. It not only simplifies the PCB
layout, but also reduces the variations of the relative distance between the power and
ground vias for all the decoupling capacitors.
Ground Via Distance (s) Angle (0)
#1 S m in 0 °
#2 V2 Sm;n 45 °
#3 S m in
00On
#4 S m in OO O 0
#5 S m in 270 °
Power Via Ground Via
Figure 3. Localized via pattern around the power via.
To represent the relative relationships between the adjacent vias, the localized via
pattern is defined to include each power via and its surrounding ground vias. To effectively
identify each via’s position, the via pattern records the distance s and angle 0 between each
ground via and the centered power via, as illustrated in Figure 3. Note that the zero-degree
14
axis can be defined from any ground via, because we only care about the relative
relationships of the vias. Based on the information in Figure 3, the distance between any
two ground vias can be further calculated. Thus, the relative positions of the included vias
can be uniquely determined by the via pattern.
Figure 4. Formulated via patterns of two categories.
Depending on the connected components, there are mainly two categories of vias
on the top and bottom layers: the IC vias and the decoupling capacitor vias. Examples of
these two categories are depicted in Figure 4.
• For the IC vias in Figure 4 (a), three patterns are formulated based on the IC pinout.
Since the pads under IC are well organized with the fixed pitch size, by applying
the via-in-pad technique, the distances between the power and ground vias can be
15
easily determined by either 1 or V2 times of the pitch size. The angles are simply
multiples of 450.
• For the decoupling capacitor vias in Figure 4 (b), only one pattern is needed to
represent all eight decoupling capacitors in the same package. This pattern is
formulated by the two vias in each capacitor’s power and ground pads, as it
provides the least impedance path for the current. Even though these capacitors are
randomly placed and the absolute via positions are difficult to determine, the
relative distance between the power and ground vias in each pattern is identical
because of the same package size.
At last, if the blind/buried vias are moved on the internal layers, the via patterns
need to be adjusted accordingly based on the formulated patterns from the top or bottom
layers.
This pattern formulation approach assumes the power and ground vias in the IC
region are alternately placed, as illustrated in Figure 4 (a). It is because the alternating via
placement achieves lower parasitic via inductance than the grouped via placement [20],
[21]. As a result, the magnetic flux is confined between the power via and its surrounding
ground vias, so the coupling among the power vias can be neglected in practical PCB
designs. The examples in Figure 4 demonstrate the flexibility of our pattern-based method
to handle the complicated PCB structures in mobile platforms. The localized via patterns
serve as the fundamental units to calculate the parasitic via inductance and resistance for
the entire PCB structure.
16
2.2. PATTERN CALCULATION
One of the most critical parasitic elements in the PDN design is the via inductance.
It is caused by the current induced magnetic flux penetrating through the loop formed by
the adjacent power and ground vias.
Figure 5 (a) shows a simple case that contains one power via carrying current I and
one ground via with the same return current. h is the via length, s is the distance between
these two vias, and r is the via radius. Since the power and ground vias are typically placed
between paralleled planes, they can be regarded as relatively long straight wires based on
the image theory (equivalent to the h >> r condition) [22]. From the Biot- Savart law for
this case, the magnetic flux density B around the power via can be calculated as:
(1)
where p is the magnetic permeability, and x is the distance to the power via. Note that the
return current flowing on the ground via equally contributes to the magnetic flux penetratin
through the loop. So the total magnetic flux O is:
Figure 5. Magnetic flux though the loop formed by adjacent vias.
17
(2)
The per-unit-length (PUL) inductance for this simple case is:
<£>(«)T (a) _PUl Ih
= t In ( S- 7r (3)
For a more complicated case, the power via may be surrounded by multiple ground
vias, as shown in Figure 5 (b). In loop 1, the induced magnetic flux of ground via 2 has an
opposite direction from that of ground via 1 and power via. It implies that the presence of
ground via 2 reduces the total magnetic flux in loop 1. With the same distance s to the
power via, the return current is evenly distributed on the two ground vias (I/2 on each
ground via). The total magnetic flux in loop 1 can be calculated as:
(4)
Symmetrically, the total magnetic flux in loop 2 is equal to the total magnetic flux in loop
1: ^ (oop 2 = ^ioop 1. Then, the PUL inductance for this case is determined by the magnetic
flux of these two paralleled loops:
(5)
The derivations of (3) and (5) demonstrate the fundamental principles to calculate
the PUL inductance based on the two simple cases. However, a real PCB structure may
contain more complicated patterns. Therefore, a generalized equation for PUL inductance
calculation needs to be developed based on an arbitrary via pattern.
18
Suppose a via pattern has N ground vias and the via radius is r, as illustrated in
Figure 6. For each ground via i where i e [1, N], its distance to the power via is si. To
maintain the law of current conservation, the total returning current I is distributed among
all the ground vias. The ground via with longer distance to the power via shares less current.
So a current coefficient ci is defined to represent the portion of the current flowing on
ground via i, where ^?L1 ci = 1. From (3), the impedance to the ground via is proportional
to the logarithm of its distance ln(s). Therefore, the current coefficient ci can be estimated
based on the inverse logarithmic relationship of the distance as:
(6)
Figure 6. Illustration of the magnetic flux for an arbitrary via pattern.
19
Similar with (2), in the loop formed by ground via i and power via, the magnetic flux
contributed by these two vias is:
(7)
The magnetic flux induced by other ground vias should also be considered. For a
different ground via j where j e [1, N] and j ^ i, its distances to the power via and ground
via i are Sj and sij, respectively. The current coefficient of ground via j is Cj. The magnetic
flux, induced by ground via j and penetrates through the loop formed by ground via i and
power via, is calculated as the integral of magnetic flux density from sij to Sj:
(8)
Note that the sign of i is determined by the ratio of Sj and sij, which implies that other
ground via’s impact on the total magnetic flux is dependent on the relative positions of the
vias.
Combining (7) and (8), the total magnetic flux in the loop formed by ground via i
and power via is:
(9)
20
Applying (9) to all the ground vias, the total magnetic flux in the N loops can be obtained,
respectively. Following the same principle of (5), the PUL inductance for the generalized
via pattern is calculated based on the magnetic flux of all the loops connected in parallel:
(10)
Another parasitic element is the via resistance, which has a secondary effect on the
PDN impedance mainly at dc and the resonant frequencies. Depending on the frequency f,
the skin depth 5 is calculated as:
(11)
where p is the resistivity and p is the permeability of the conductor material. For a
standard-size via when the via radius r >> 5, the effective conduction area Aeff can be
determined using a simplified equation: Aeff = n [ r2 — (r — 5)2]. However, the dimension
of the micro-via used in mobile platform PCB is comparable with the skin depth within our
frequency range of interest (300 kHz ~ 300 MHz). Hence, a more accurate method using
the “truncated exponential decay” approach [23] is used to derive the modified skin depth
5' for the via resistance calculation:
21
(12)
The asymptotically correct formula for the effective area is:
A i f = n [ 2 r S ' - S ' 2 ] • (1 + y ) , (13)
where (1 + y) is a divisor correction coefficient based on a Modified Lorentzian function
y. The fitted expression of y is also provided in [23]:
(14)
Finally, the PUL resistance for a single via is calculated as:
(15)
Figure 7. Arbitrary via patterns for validation.
22
The pattern-based analytical equation (10) can be applied to an arbitrary via pattern.
Here a commercial RF simulation tool, ANSYS 2D Extractor, was used to validate this
equation for various via patterns and distances. As illustrated in Figure 7, seven patterns
were built in the simulation tool with different numbers and positions of the ground vias.
The via diameter was 0.075 mm. The minimum via distance smin was sweeping from 0.3
mm to 0.8 mm for all the patterns. These values were chosen based on the typical
specification of the HDI PCB in mobile platforms. The simulated PUL via inductance,
obtained using the ANSYS 2D Extractor, and the calculated PUL via inductance, obtained
based on the proposed analytical equation, are compared in Figure 8. Good correlations are
observed for all the cases where the largest difference between the calculated and simulated
PUL inductances is 9.54 pH/mm (with a maximum discrepancy of 1.1%). It verifies the
accuracy of our proposed analytical equation for an arbitrary via pattern. Equation (15) for
the PUL via resistance is validated in Section 4.2.
Since the pattern is a 2-D structure, the vias are equivalently treated as relatively
long wires during the calculation of the PUL parasitic elements. Therefore, the edge-effect,
such as the fringing fields at the end of the vias or the proximity to the plane edges, may
affect the accuracy of our method. To investigate the impact of the edge-effect, another 3
D electromagnetic field simulation tool, CST Studio, was used to simulate the via
inductance of different lengths. “Pattern 1”, “Pattern 5”, and “Pattern 7” in Figure 7 were
evaluated to check the smallest, median, and largest PUL inductance cases. The minimum
via distance s min was fixed to 0.35 mm and the plane size was 2 mm by 2 mm. As shown
in Figure 9, a port was added to the power via, the ground via(s) were shorted to provide a
2.3. SIMULATION STUDIES ON ACCURACY AND EDGE-EFFECT
23
return path for the injected current. The minimum via distance s min was fixed to 0.35 mm.
The via length varied from 0.02 mm to 0.10 mm. The edge-effect causes additional
parasitic PUL inductance when the via length is smaller than 0.04 mm. For a mobile
platform PCB, the typical dielectric thickness between layers is 0.045 ~ 0.060 mm. In other
words, the error introduced by the edge-effect is negligible in this application. The
modeling of the edge-effect is beyond the scope of this article.
^ P a tte rn 3 < ► P a tte rn 2
■i T P a tte rn 7
1400InductanceCalculated PULLine
Simulated InductanceD o ts PUL P a tte rn 11200
1000P a tte rn 6
800P a tte rn 4P a tte rn 5
400
M inimum Via Distance (mm)
Figure 8. Comparison of via inductance for different patterns and distances.
24
3. PDN MODELING METHODOLOGY
With the assistance of this pattern-based analytical method, a practical modeling
methodology is proposed to calculate the PDN impedance of the entire PCB. The flow
diagram is depicted in Figure 10. On the left side of this diagram, the “Via Pattern
Database” indicates that the via patterns from the previous projects can be stored and
adapted for future designs. It is not only useful to determine the impedance of a known
PCB structure, but also helpful to improve the layout especially in the pre-design stage.
The dashed arrow lines represent the interactions with this database. On the right side of
this diagram, the main workflow is divided into four steps. The first three steps, including
“(1) PCB division”, “(2) pattern identification”, and “(3) circuit reconstruction”, construct
25
an equivalent circuit with one-to-one correspondences to the PCB physical geometry.
Then, the PDN impedance can be obtained from the circuit simulations. If the target
impedance is not satisfied for the initial design, an iteration process is conducted to further
improve the impedance in the “(4) design optimization” step.
Update pattern or number
Iterationprocess
Satisfy Target Impedance?
StartAdd new orload existingvia patternsVia Pattern (1) PCB DivisionDatabase
DetermineLoop up existingvia patterns regions
(4) DesignOptimization
(2) PatternIdentification
Calculate PULelements
(3) CircuitReconstruction
Get equivalentcircuits
NO
YES
End
Figure 10. Flow diagram of the modeling methodology.
26
Following the modeling methodology, a real mobile phone PCB, illustrated in
Figure 11, is investigated as an example. Among the total 12 layers, the 7th layer is the
main power plane connecting all the decoupling capacitors and IC, and the 3rd layer is the
sub power plane to unify the voltage under the IC. The ground planes are located on the
2nd, 4th, 6th, 9th, and 11th layers. The copper thickness is 0.023 mm for the top and bottom
layers, and 0.018 mm for all the internal layers. There are 0402 package (22 pF, 10 pF)
and 0201 package (2.2 pF) decoupling capacitors placed on both the top and bottom layers.
All of the 0201 package capacitors are placed directly under the IC, while the other 0402
package capacitors are placed besides the IC due to the space limitation.
D i e l e c t r i c
T h i c k n e s s
0 .0 4 5 m m
0 .0 4 5 m m
0 .0 5 0 m m
0 .0 5 0 m m
0 .0 4 5 m m
0 .0 6 0 m m
0 .0 4 5 m m
0 .0 5 0 m m
0 .0 5 0 m m
0 .0 4 5 m m
0 .0 4 5 m m
Figure 11. Physical geometry and region division of the mobile phone PCB.
27
Figure 12. The equivalent circuit corresponding to the PCB under investigation.
3.1. PCB DIVISION
Seen at the input port of the load IC, the current in the PDN is always flowing along
the power vias from the decoupling capacitors and returning along the nearby ground vias.
By tracing the current flow path, the multilayered PCB structure can be divided vertically
into different regions between the main power plane and the ground plane. For the target
PCB in Figure 11, there are five regions associated with the physical geometry. “Region
1” and “Region 2” correspond to the vias directly under the IC from the top and bottom
layers to the main power plane, respectively. “Region 2” also connects to the multiple 2.2
pF decoupling capacitors on the bottom layer. “Region 3”, “Region 4”, and “Region 5”
correspond to the vias between the 0402 package decoupling capacitors and the main power
plane. Since these capacitors are located besides the IC, we should include the plane
inductance to reflect the current distribution on the main power plane. The mutual
inductance among different regions are negligible because the distance between regions is
28
relatively larger than the distance from the power via to the nearby ground vias [24]. Within
each region, the parasitic elements of the vias can be merged into the equivalent inductance
and resistance. The equivalent circuit of the entire PCB structure is depicted in Figure 12.
3.2. PATTERN IDENTIFICATION
The pattern formulations for the IC and decoupling capacitor vias are introduced in
Section 2. Accordingly, the patterns in “Region 1” are identified based on the IC pinout,
and the patterns in other regions are identified based on the packages of the decoupling
capacitors. Furthermore, an optimal pattern size needs to be selected for practical purposes.
It is because a larger pattern includes more nearby ground vias, which leads to a more
accurate result but also increases the complexity in pattern formulation. For the IC region
shown in Figure 13, the pitch size s min is fixed to 0.35 mm and the radius of the effective
pattern area r eff is investigated using 1, V2, 2, and V5 times smin. When r eff = smin, only
two ground vias are included in this pattern, but it yields a falsely high PUL via inductance
of 635 pH/mm. By increasing reff, the PUL via inductance converges to 537 pH/mm. The
optimal pattern size is achieved when r eff = 2smin as it provides an accurate result with
the simplest structure. This experimental-based conclusion can be verified using other via
patterns. Therefore, it is suggested to include the ground vias whose relative distance is
within two pitch sizes into the pattern.
29
Effective pattern areaRadius rcIT
Pitch size smin
O Power Via
0 Ground Via
S ' 7 0 0
I 6003 5 0 0
o 4 0 0
% 3 0 0
| 200
3 100A
■ C a lc u la te d 0 S im u la te d
Figure 13. Selection of the optimal pattern size.
3.3. CIRCUIT RECONSTRUCTION
This step aims to determine all the components of the equivalent circuit in Figure
12. The PUL via inductance and resistance of the identified patterns can be calculated based
on the analytical equations (10) and (15). Then, the inductance of a via segment is obtained
by multiplying the PUL inductance by the effective length of this segment, and the
resistance is obtained by multiplying the PUL resistance by the full segment length.
Compared with the full length, the effective length excludes the layer thickness of the
power and ground planes. It is because the inductance is related to the loop formed by the
adjacent vias. The signal planes can also be ignored since they do not affect the impedance
of the power nets. As illustrated in Figure 14, the inductance and resistance of multiple
segments in the same region can be merged depending on their series or parallel
relationship. As a result, one equivalent inductance and one resistance are generated to
represent the parasitic elements for each region.
30
The plane capacitance is insignificant for the mobile phone PCB because of the
limited layout space, so the simple parallel-plate capacitor equation is sufficient to estimate
the plane capacitance:
(16)
where s is the dielectric permittivity, A is the plane area, and d is the dielectric thickness
between the plane pair. The total inductance between the main power plane pair (“Plane
inductance” in Figure 12) is obtained based on the cavity model [15]—[18]. By merging the
vias of the same region, the cavity model can be simplified by setting one port located at
the center of each region instead of identifying the absolute positions for all the vias. At
last, “Region 2”, “Region 3”, “Region 4”, and “Region 5” connect to the decoupling
capacitors on the top and bottom layers. For high accuracy, the SPICE model of the
decoupling capacitor is employed in the circuit simulation since it includes both the
parasitics, such as the equivalent series inductance/resistance (ESL/ESR) and the derating
effects. Similar with the via inductance and resistance, the decoupling capacitors in the
same region are in parallel and can be merged together.
31
3.4. DESIGN OPTIMIZATION
With the constructed equivalent circuit, the PDN impedance can be assessed by the
circuit simulations. If the target impedance is not satisfied, an iteration process is conducted
to further optimize the PDN design. The divided regions are helpful to identify the most
critical component contributing to the total impedance. A common optimization method is
to apply more decoupling capacitors, which can be easily implemented as increasing the
number of the decoupling capacitor via pattern in our modeling methodology. Adding more
ground vias is another feasible solution to reduce the parasitic via inductance, which is
equivalent to changing the via pattern in design. The proposed pattern-based analytical
equation is capable of predicting the PUL via inductance accurately, in order to provide
quantitative guidance for the system designer. For instance, Figure 8 shows the PUL
inductance with different via distances of the seven patterns in Figure 7. According to the
calculated results, the designer is able to achieve lower impedance level with better
utilization of the layout space for the new design. So our modeling methodology is
especially useful in the pre-design stage when the PCB layout is not fully completed.
4. VALIDATION
The PDN impedance of the mobile phone PCB in Figure 11 was measured to
validate our modeling methodology. The shunt-thru method was used to achieve the
accurate mO impedance measurement [25], as depicted in Figure 15. By landing the two
micro-probes (PacketMicro GR201504) between the IC’s power and ground pads, the
transmission coefficient S21 was measured by the vector network analyzer (Keysight
32
E5061B). Then, the impedance o f the entire PCB structure can be derived from the
measured S-parameter: Zpdn = 25 • S2i(1 — S21). To avoid the mutual coupling between
the micro-probes, the landing locations were separated at the opposite sides o f the IC
pinout. Full 2-port calibration and port extension were applied to move the reference planes
to the micro-probe tips. Ferrite cores were attached to the test cable to eliminate the
measurement error caused by the cable ground loop [26]. To include the derating effects
of the decoupling capacitors, 0.75 V dc bias voltage was enabled in the vector network
analyzer. The frequency range in this measurement was from 300 kHz to 300 MHz.
V e c to r n e tw o r k
a n a ly z e r (V N A )
F e r r it e c o r e
E n a b le
d c b ia s T a r g e t P C B
M ic r o -p r o b e
C a lib r a t io nP o r t e x te n s io n
Figure 15. Photograph and diagram of the shunt-thru measurement setup.
33
The layout of this mobile phone PCB was also imported into a board-level
electromagnetic field solver, Cadence Sigrity PowerSI. In addition to the total PDN
impedance, the parasitic inductance and resistance of each region were simulated by
manually setting the ports at different locations and layers. As a result, the accuracy of our
modeling methodology can be validated by both whole structure and region-by-region
comparisons.
4.1. WHOLE STRUCTURE COMPARISON
The impedance for the whole structure is contributed by all the components
embedded in the PCB. However, different components have dominant effects at different
frequencies. For example, capacitance usually dominates at a lower frequency and the via
inductance is typically critical at a higher frequency. Therefore, it is desired to achieve
good correlations with the simulations and measurements for the entire frequency range.
The calculated result based on our modeling methodology was using one lumped
port grouping all the IC vias on the top layer. However, it is unrealistic to implement this
lumped port in the measurement setup. As discussed above, the measured result was
converted from the 2-port S-parameter, which resulted in lower IC via inductance and
caused lower impedance at higher frequencies. So the calculated and measured results
could not be compared directly. To solve this issue, two simulated results obtained using
the Cadence Sigrity PowerSI were added with the lumped port and 2-port configurations,
respectively. Then, the four types of results could be divided into two groups: the lumped
port group including the calculated and simulated results; and the 2-port group including
the simulated and measured results.
34
Two criteria are evaluated to validate our modeling method: the resonant
frequencies and the magnitude of the PDN impedance. As shown in Figure 16, the
impedance curves in the same configuration group agree well with each other. The three
resonant frequencies, located at 1.5 MHz, 2.0 MHz, and 5.5 MHz, are all accurately
captured with less than 100 KHz difference. The maximum variations of the impedance
magnitude are within 0.26 mO and 0.33 mO for the lumped and 2-port configurations,
respectively. Between the two groups, we can indirectly compare the calculated and
measured results with respect to the difference of the two simulated impedance. The good
correlations across the entire frequency range validates the accuracy of our modeling
methodology for the whole PCB structure.
Calculated : Lumped PortSimulated : Lumped PortSimulated : 2-PortMeasured : 2-Port
R e d u c e d in d u c ta n c e
c a u s e d b y 2 - p o r t s e tu p
Frequency (MHz)
Figure 16. Comparison of the PDN impedance for the whole PCB structure.
35
Table 1. Region-by-region comparison of parasitic elements.
Location Parasitic Calculation SimulationR eg ion 1
L a y e r 1 ~ 3
In d u c tan ce 2 .10 p H 2 .39 pH
R esis tan ce 0 .05 0 .06 m fJ
R eg ion 1
L ay e r 3 ^ 7
In d u c tan ce 4.81 p H 5.48 pH
R esis tan ce 0 .14 0 .14 m£2
R eg ion 2
L a y e r 7 ^ 1 2
Inductance 13.51 pH 12.93 pH
R esis tan ce 0 .20 m£2 0 .19 m i l
R eg ion 3
L a y e r 1 ~ 6
In d u c tan ce 289 pH 285 pH
R esis tan ce 2.95 4 .10
R eg ion 4
L a y e r 7 ~ 1 2
In d u c tan ce 312 pH 312 pH
R esis tan ce 3 .62 mS2 4 .00 mf2
R eg ion 5
L a y e r 7 ^ 1 2
Inductance 242 pH 240 pH
R esis tan ce 3 .62 m£2 4 .0 0 m i l
4.2. REGION-BY-REGION-COMPARISON
Since the proposed modeling methodology is built on the basis of the divided
regions, it is important to validate the equivalent inductance and resistance for each region.
The calculated and simulated parasitic elements for different regions are compared in Table
1. “Region 1” is split into two parts because of the sub power plane on the layer 3. Other
regions have only one set of equivalent inductance and resistance. The via resistance are
obtained at the 6 MHz resonant frequency.
The error between the calculated and simulated parasitic inductance is within
12.2% for the IC vias and within 1.4% for the decoupling capacitor vias. The good
agreement shows this methodology can not only estimate the via inductance accurately,
but also identify the critical components contributing to the total PDN impedance. The
calculated parasitic resistance generally correlate well with the simulated results (with a
36
discrepancy of 9.5%), except for “Region 3” (with a discrepancy of 28.0%). It is because
there are multiple paralleled vias laid closely in the same pad of the decoupling capacitor
in this region. Thus, the increased proximity effect causes higher ac resistance. However,
the impedance curve is not significantly affected since the via resistance has a secondary
impact mainly at dc and the resonant frequencies. So the current accuracy level for the
resistance calculation is acceptable. The proximity effect modeling is not the focus of this
article and will be discussed in future studies.
5. CONCLUSION
In this article, the limitations of the existing PDN modeling methods are discussed
with the focus on mobile platform applications. To overcome these issues, a novel pattern-
based analytical method is proposed for the impedance calculation. By utilizing the relative
relationships between the adjacent vias, the localized via patterns are formulated based on
the physical geometry. Then, the parasitic via inductance and resistance are analytically
derived for arbitrary via patterns. It can be shown that our method has better flexibility to
handle the complicated PCB structure in mobile platforms. In addition, a practical
modeling methodology is developed to model and guide the PDN design. With the
capability of accurate and layout-free impedance estimation, this methodology is especially
useful for the pre-design stage to accelerate the development process. Finally, the proposed
analytical equations and the modeling methodology have been extensively validated
through the measurements and simulations on a real mobile phone PCB. Good agreements
can be observed from both whole structure and region-by-region comparisons.
37
REFERENCES
[1] M. Swaminathan, J. Kim, I. Novak, and J. P. Libous, “Power distribution networks for system-on-package: Status and challenges,” IEEE Transactions on Advanced Packaging, vol. 27, no. 2, pp. 286-300, 2004.
[2] W. D. Becker, J. Eckhardt, R. W. Frech, G. A. Katopis, E. Klink, M. F. McAllister, T. G. McNamara, P. Muench, S. R. Richter, and H. Smith, “Modeling, simulation, and measurement of mid-frequency simultaneous switching noise in computer systems,” IEEE Transactions on Components, Packaging, and Manufacturing Technology: Part B, vol. 21, no. 2, pp. 157-163, 1998.
[3] X. Zhou, P.-L. Wong, P. Xu, F. C. Lee, and A. Q. Huang, “Investigation of candidate vrm topologies for future microprocessors,” IEEE Transactions on Power Electronics, vol. 15, no. 6, pp. 1172-1182, 2000.
[4] M. Halpern, Y. Zhu, and V. J. Reddi, “Mobile CPU’s rise to power: Quantifying the impact of generational mobile cpu design trends on performance, energy, and user satisfaction,” in 2016 IEEE International Symposium on High Performance Computer Architecture (HPCA). IEEE, 2016, pp. 64-76.
[5] L. D. Smith, R. E. Anderson, D. W. Forehand, T. J. Pelc, and T. Roy, “Power distribution system design methodology and capacitor selection for modern cmos technology,” IEEE Transactions on Advanced Packaging, vol. 22, no. 3, pp. 284291, 1999.
[6] I. Erdin and M. Nakhla, “Mixed circuit/electromagnetic analysis of field coupling to high speed interconnects in inhomogenous medium,” in 1999 IEEE International Symposium on Electromagnetic Compatability. Symposium Record (Cat. No. 99CH36261), vol. 1. IEEE, 1999, pp. 446-449.
[7] M. Feliziani and F. Maradei, “Circuit-oriented fem: Solution of circuit-field coupled problems by circuit equations,” IEEE Transactions on Magnetics, vol. 38, no. 2, pp. 965-968, 2002.
[8] E.-X. Liu, X. Wei, Z. Z. Oo, E.-P. Li, and L.-W. Li, “Modeling of advanced multilayered packages with multiple vias and finite ground planes,” in 2007 IEEE Electrical Performance of Electronic Packaging. IEEE, 2007, pp. 275-278.
[9] A. E. Ruehli, “Inductance calculations in a complex integrated circuit environment,” IBM journal of research and development, vol. 16, no. 5, pp. 470481, 1972.
38
[10] A. E. Ruehli and P. A. Brennan, “Efficient capacitance calculations for threedimensional multiconductor systems,” IEEE Transactions on Microwave Theory and Techniques, vol. 21, no. 2, pp. 76-82, 1973.
[11] A. E. Ruehli, “Equivalent circuit models for three-dimensional multiconductor systems,” IEEE Transactions on Microwave Theory and Techniques, vol. 22, no. 3, pp. 216-221, 1974.
[12] P. Johns, “The art of modelling,” Electronics & Power, vol. 25, no. 8, pp. 565-569, 1979.
[13] W. J. Hoefer, “The transmission-line matrix method-theory and applications,” IEEE Transactions on Microwave Theory and Techniques, vol. 33, no. 10, pp. 882893, 1985.
[14] K. Lee and A. Barber, “Modeling and analysis of multichip module power supply planes,” IEEE Transactions on Components, Packaging, and Manufacturing Technology: Part B, vol. 18, no. 4, pp. 628-639, 1995.
[15] J. Kim, L. Ren, and J. Fan, “Physics-based inductance extraction for via arrays in parallel planes for power distribution network design,” IEEE Transactions on Microwave Theory and Techniques, vol. 58, no. 9, pp. 2434-2447, 2010.
[16] J. Kim, J. Fan, A. E. Ruehli, J. Kim, and J. L. Drewniak, “Inductance calculations for plane-pair area fills with vias in a power distribution network using a cavity model and partial inductances,” IEEE Transactions on Microwave Theory and Techniques, vol. 59, no. 8, pp. 1909-1924, 2011.
[17] J. Kim, K. Shringarpure, J. Fan, J. Kim, and J. L. Drewniak, “Equivalent circuit model for power bus design in multi-layer pcbs with via arrays,” IEEE Microwave and Wireless Components Letters, vol. 21, no. 2, pp. 62-64, 2011.
[18] K. Shringarpure, S. Pan, J. Kim, B. Achkir, B. Archambeault, J. Drewniak, and J. Fan, “Formulation and network reduction to a physics based model for analysis of the power distribution network in a production level multi-layered printed circuit board,” IEEE Transactions on Electromagnetic Compatibility, vol. 58, no. 3, pp. 849-858, 2016.
[19] G. Cecere, N. Corrocher, and R. D. Battaglia, “Innovation and competition in the smartphone industry: Is there a dominant design?” Telecommunications Policy, vol. 39, no. 3-4, pp. 162-175, 2015.
[20] K. Shringarpure, “Printed circuit board power distribution network modeling, analysis and design, and, statistical crosstalk analysis for high speed digital links,” 2015.
39
[21] B. Archambeault, B. Zhao, K. Shringapure, and J. Drewniak, “Design tips,” IEEE Electromagnetic Compatibility Magazine, vol. 4, no. 2, pp. 106-107, 2015.
[22] T. H. Hubing, T. P. Van Doren, and J. L. Drewniak, “Identifying and quantifying printed circuit board inductance,” in Proceedings of IEEE Symposium on Electromagnetic Compatibility. IEEE, 1994, pp. 205-208.
[23] D. W. Knight, “Practical continuous functions for the internal impedance of solid cylindrical conductors,” G3YNH info, 2016. [Online]. Available: http://www. g3ynh.info/zdocs/comps/Zint.pdf.
[24] B. Zhao, “A physics-based approach for power integrity in multi-layered pcbs,” Master’s thesis, Missouri University of Science and Technology, 2017.
[25] Keysight. (2014) Evaluating dc-dc converters and pdn with the e5061b lf-rf network analyzer. [Online]. Available: http://literature.cdn.keysight.com/litweb/ pdf75990-5902EN.pdf.
[26] I. Novak, Y. Mori, and M. Resso, “Accuracy improvements of pdn impedance measurements in the low to middle frequency range,” Proc. DesignCon 2010, pp. 1-4, 2010.
40
II. TOPOLOGY-BASED ACCURATE MODELING OF CURRENT-MODE VOLTAGE REGULATOR MODULES FOR POWER DISTRIBUTION
NETWORK DESIGN
ABSTRACT
Power distribution network (PDN) is essential in electronic systems to provide
reliable power for load devices. Thus, modeling of PDNs in printed circuit board and
package has been extensively studied in the past few decades. However, the voltage
regulator module (VRM), which serves as the ultimate voltage source, was not thoroughly
investigated. The conventional VRM model is either represented by a simple resistor-
inductor equivalent circuit or implemented with simple assumptions. It lacks of accuracy
for modern VRM devices. In this work, a generic behavior model, including both the power
stage and control loops, is developed for the current-mode buck VRM. A novel method is
also proposed to synthesize the continuous and discontinuous conduction modes for
transient load responses. Through the measurement-based characterization, the model
parameters are optimized to match with the actual design. Furthermore, this model can be
applied to both the time-domain and frequency-domain circuit simulations to predict the
voltage droop and output impedance, respectively. The accuracy of the model is validated
using an evaluation board containing the single-phase and multiphase VRMs. It is
demonstrated that the proposed model for VRM with control loops can be easily integrated
into the PDN analysis and optimization.
41
1. INTRODUCTION
Power distribution network (PDN) is essential in electronic systems to provide
reliable power for load devices [1]. The primary objective of the PDN design is to minimize
the voltage fluctuations caused by the load current in order to maintain normal integrated
circuit (IC) functions [2]. As depicted in Figure 1, a typical PDN consists of four main
parts: voltage regulator module (VRM), printed circuit board (PCB)-level PDN, package-
level PDN, and on-die PDN. When the transient current draws at the die bumps, a voltage
droop may occur and affect the system’s stability. The voltage droop in the time-domain
and the PDN impedance in the frequency-domain are tightly related with each other, as
shown in Figure 2. The high frequency response is mainly determined by the package/die
parasitics (usually above 100 MHz) or the PCB design (usually 300 kHz ~ 100 MHz), and
the low frequency response is typically dominated by the VRM (usually below 300 kHz).
However, driven by ambitious user demands, modern ICs are dealing with more
sophisticated scenarios while the voltage tolerance margin becomes progressively tighter
[3]. To compensate for a large droop, the voltage sensing location is preferred to be closer
to or even directly at the die bumps, so the voltage regulation (VR) control gains an
increased open-loop bandwidth to react to faster transient events. This can introduce more
coupling between the VR control and the PDN. An extreme case is the in-package
integrated voltage regulator, where the control loop directly affects the first droop. Thus,
influence of the VRM on the PDN design becomes increasingly more significant,
especially in the high switching frequency and wide bandwidth applications.
42
Figure 1. Main elements in the end-to-end PDN.
In the past few decades, the modeling of PDNs in PCB and package has been
extensively studied. The fast transmission-line method (TLM) [4] and the cavity model
method [5], [6] were successfully applied to the multilayered PCB. The chip-package
hierarchical PDN structures were modeled by the partial-element equivalent circuit (PEEC)
[7] and the decomposed segmentation method [8]. In addition, the full-wave
electromagnetic modeling methods, such as the finite-difference time-domain (FDTD) [9],
the finite-element method (FEM) [10], and the method-of-moments (MOM) [11], were
also commonly used to simulate the PCB and package PDN impedance. For VRM
modeling, conventionally the first-order linear resistor-inductor (RL) or the second-order
4-element RL circuits were fitted to match the VRM output impedance [12]. With the
evolution in technologies, VRMs are integrated with the controller to effectively minimize
voltage deviations caused by load transients [13]. As a result, discrete passive elements are
incapable of representing the VRM’s control loop responses. A behavior model including
both the switching part and the feedback loop was proposed in [14]. However, the feedback
loop was inaccurately simplified as a boost converter. Ahmadi et al. developed closed-loop
43
mathematical equations to calculate the output impedance of dc-dc switching converters
[15]. But this approach was purely mathematical, and was not intuitive to provide insights
to improve the VRM design. In [16], an encrypted device model provided by the IC vendor
was employed for VRM transient simulations. Even though the simulation model was
tuned to correlate with the actual performance, it was only applicable to a specific device
from this vendor and the detailed implementations were not accessible due to the
confidential concerns. In sum, despite the various PCB-level and package-level PDN
modeling methods, there lacks an effective VRM model for PDN designs. Additionally,
the higher the integration level of the VRM, the more complicated the internal design,
which is mostly protected or encrypted by the IP-cores, will be. Thus, it is desired to
develop a behavior model to bypass the IP-protection concerns.
0 .85
0 .75
0 .65
------V o lta g e a t d ie b u m p■■■■ V o lta g e a t V R M o u tp u t
£ 10'1
0L o w F r e q u e n c y R e s p o n s e
H ig h r r e q u e n c y R e s p o n s e
2 3 4T im e (/j, s)
(a) Time domain voltage droop
P a c
ii
PCB |
ca g e
l ,1! D ie 1
V R M I I I111h 111 \ / 11 V f 1
1111
V / 1 \ / 1 \ / 1 \ / 1 \ / 1 \ a 1 \ / 1 V M 1
\ / 1
ill kli \I1 \I1 \ I1 \ I1li|ili
\ / 1 W i i -
10-
F re q u e n c y (H z)
(b) Frequency domain impedance
Figure 2. System-level PDN performance.
44
In this paper, a topology-based VRM model is proposed to address these
challenges. By comparing different topologies in Section 2, this work mainly focuses on
the widely used current-mode buck VRM. In Section 3, a generic behavior model based on
the current mode topology is established including equivalent key components in the
voltage control loop, the current control loop, and the power stage. Measurements are
utilized in Section 4 to characterize the model parameters and also validate the proposed
model using an evaluation board (EVB) which contains both single-phase and 3-phase
VRMs. Section 5 concludes this paper.
2. VRM TOPOLOGY
VRM may refer to a wide range of chips and dc-dc converters with controllers. The
step-down dc-dc converter for high-current rails that are especially sensitive to the transient
voltage droop is studied herein because it is closely related to the PDN performance. Its
power stage is essentially a buck converter. As depicted in Figure 3, the input voltage V in
is regulated by the high-side and low-side MOSFETs. The gate driver, which includes an
oscillator, an RS flip flop, and a drive logic circuit, controls the alternating on-off states of
the MOSFETs with the switching frequency fsw and the duty cycle d. The switching node
is connected to the output inductor L and the smoothing capacitor Co u t, in order to provide
the output voltage Vo u t. Depending on whether the inductor current iL reaches zero during
each switching cycle, the converter may operate in the continuous or discontinuous
conduction mode (CCM or DCM). Since the CCM and DCM correspond to the heavy and
light load conditions, respectively, it is crucial to model both modes to accurately predict
45
the transient load response. To minimize the voltage fluctuation caused by the IR drop and
ac noise, the VRM typically integrates feedback control loops. Figure 4 demonstrates two
types of the control topologies: the voltage-mode and the current-mode [17].
Figure 3. Diagram of an open-loop dc-dc buck converter.
out o u tG a te G a te
D r iv e r D r iv e r
U I'-sen' &i
e rr err
A V H Vref
(a ) V o lta g e -M o d e (b ) C u rre n t-M o d e
Figure 4. Typical feedback control topologies in VRM.
46
For the voltage-mode in Figure 4 (a), the feedback voltage Vfb can be sensed either
at the output voltage node or remotely at the die bumps. Then, it is compared with an
internal reference voltage Vr e f, which is precise and stable over the temperature. The error
voltage Verr, calculated by subtracting Vfb from Vr e f, is further amplified by the error
amplifier (EA) and the compensation circuits (n, C 1 , and C2). The compensated voltage Vc
is then sent to the control input. The duty cycle d as the control variable can be determined
based on the input level of Vc and a modulated saw-tooth signal. As a result, the power
stage reacts to reduce V err as much as possible. However, the voltage-mode control has
several notable drawbacks. First, the loop gain is limited and proportional to V in . In
addition, the changes in load must be sensed by Vfb in order to be corrected by the feedback
loop, which results in a slow response. At last, it usually requires a complicated Type III
compensation [17] and the CCM/DCM differences also add challenges for the
compensation circuit.
To solve these drawbacks of the voltage-mode control, the current-mode control is
developed and widely adopted in industrial applications. As shown in Figure 4 (b), the
current-mode control introduces an additional feedback loop from iL . The total current
sensing gain is R i, which is comprised of the sensing resistor Rsen and the current
amplifier’s gain ai . The compensated voltage Vc from the voltage feedback loop is then
compared with the voltage converted from iL , and initiates the duty cycle d for the next
switching period. When d > 50%, the slope compensation is also required for the current
feedback loop to overcome the instability caused by the sub-harmonics oscillations [18].
The current-mode control can address the slow response issue of the voltage-mode control
because iL responds immediately to load changes. Furthermore, this system with a single
47
pole roll-off of the control-to-output transfer function can be stabilized with a simpler Type
II compensation circuit around the EA. The circuit also exhibits a higher gain bandwidth
compared to the voltage-mode control [19]. In sum, the current-mode control is a superior
approach for fast dynamic responses and is widely used in the modern VRM designs.
However, this topology also possesses a larger complexity due to the dual feedback loops
and slope compensation.
3. TOPOLOGY-BASED GENERIC BEHAVIOR MODEL
In this section, a topology-based generic behavior model is developed for the
current-mode buck VRM. This model includes all the equivalent key components in the
voltage control loop, the current control loop, and the power stage based on the topology
shown in Figure 4 (b). A novel unified model is also proposed for the current control loop
to handle both the CCM and DCM load conditions. At last, the implementation of this
VRM behavior model is demonstrated in a circuit simulator, which is able to provide both
the time-domain and frequency-domain simulation results.
3.1. VOLTAGE CONTROL LOOP
The voltage control loop is fed by the feedback voltage Vfb , as shown in Figure 5
(a). The error voltage Verr is obtained as:
Verr = K - e f — Vfb. ( 1)
Since the EA is an equivalent voltage-controlled current source with the transconductance
gm , the current flowing into the compensation circuit is ic = gm V e rr. It is convenient to model
48
the Type II compensation circuit using the s-domain analysis, because the compensated
voltage Vc can be expressed as the multiplication o f ic and the EA output impedance:
(2)
where r 1 , C 1 , and C2 are the resistor and capacitors in this compensation circuit.
Choosing Verr as the input signal and Vc as the output, the transfer function o f this
system can be derived from (2):
Vc gm(s C in + 1)(3)
K e r r S2C 1C2r 1 + s ( C 1 + C 2)'
Since C 1 is set to cancel the low-frequency pole of the current-mode topology and C2 adds
a high-frequency pole to filter out the high-frequency noises, the relationship Cx » C2 is
generally satisfied in practical VRM designs. Therefore, (3) can be further simplified to:
(4)
The right-hand-side (RHS) of (4) contains a low-pass filter (LPF) and a proportional-
integral (PI) control. The first term in the RHS represents the LPF with the cut-off
frequency fc:
(5)
The second term in the RHS includes a proportional process with the coefficient kp and a
integral process with the coefficient ki :
49
(6)
(7)
In addition, the integrator has a dc gain limit kdc in real devices, which can be modeled as:
(8)
Based on (4), the equivalent model of the voltage control loop is depicted in Figure
5 (b). The mapping relationships between the topology and the equivalent model are
constructed by (5), (6), (7), and (8). Another parameter in the voltage control loop is the
control delay td from the instant when Vout is sensed to the instant when Vc is updated,
which is included in the model implementation procedure described later.
Figure 5. Modeling of the voltage control loop.
50
3.2. CURRENT CONTROL LOOP
Figure 6 illustrates the inductor current iL under the CCM and DCM load
conditions. Within each switching period Ts = 1/f, the averaged inductor current is denoted
as IL. The rising and falling slopes o f iL are denoted as Sr and Sf, respectively. Since the
slopes are related to the charging and discharging behaviors o f the output inductor, Sr and
Sf are calculated as:
q __ U n '^ (^ o n .H “1“ r L ) W iltJ , (9)
q __ 2l ( ^ oii,L “1“ r L ) W u t ( 10)
where L and rL are the inductance and equivalent series resistance (ESR) o f the output
inductor, ron ,H and ro n ,L are the on-resistance o f the high-side and low-side MOSFETs,
respectively. With the slope compensation added to the compensated voltage, the peak
inductor current iL ,p e a k is calculated as:
* L ,p e a k — n ( W dVrp ) ,rt\ (11)
where V rp is the ramp voltage for the slope compensation.
For the CCM in Figure 6 (a), since iL does not reach zero during each switching
cycle, the derivative o f IL between the adjacent cycles is determined as:
(12)
Then, the averaged inductor current for the CCM is derived based on simple figure
calculations:
— ^L,peak 2 dTs | STm -
(13)
51
Substitute (12) into (13) and move all the terms to the left-hand-side o f the equation:
*L.peak - k - ^d( 1 - d)Ts(Sv - Sf) = 0 . (14)
Define AS = Sr — Sf , then combine (11) and (14), a quadratic equation with the
independent variable d can be constructed as:
\ T‘ A S j ‘ - g r . A S + V ( | - *L I = 0. (15)
Solve (15) to obtain d in the CCM. It should be noted that the range of the duty cycle is 1
^ d ^ 0, so the solution o f d > 1 is dropped. Thus, the valid d can be expressed as:
(16)
Figure 6. The inductor current in one switching period.
52
For the DCM in Figure 6 (b), iL reaches zero within the switching cycle, so the
derivative of iL between the adjacent cycles is also zero. Suppose the ratio of the time
period for iL > 0 is dc, the relationship between d and dc is:
(17)
Similarly, the averaged inductor current for the DCM is:
%L = ~ S rdT&dc (18)
Combine (17) and (18) to solve for the two unknown parameters dc and d in the DCM:
(19)
(20)
It is clear that the current control loop is difficult to be implemented in circuit
simulations, because the CCM and DCM use different equations, (16) and (20), to
determine d for the power stage. To solve the problems, (19) is further expanded for any
sensed iL as:
(21)
Then, dc is utilized to transform the DCM iL to the iL of the special boundary case between
the CCM and DCM. As shown in Figure 6 (b), the boundary case occurs when i iL reaches
zero exactly at Ts with the same slopes. The superscript ( x ' ) is introduced to the
corresponding variables for the boundary case as:
53
V ' = *c dc (22)
? _ 6.*L dj •
(23)
Note that (16) is also applicable for the boundary case, so its duty cycle based on VC and iL
is:
(24)
At last, the DCM duty cycle can be obtained by the reverse-transformation using (21) and
(24) instead of (20) as:
d = d'dc. (25)
When the system works in the CCM, dc = 1 and (25) provides the same result as (16).
Therefore, a unified procedure is developed for the current control loop for different load
conditions, which significantly simplifies the model implementation procedure.
3.3. POWER STAGE
As depicted in Figure 3, the control variable d from the feedback control loops is
fed to the power stage to adjust the output. In this model, we mainly focus on the averaged
behavior of the VRM to predict the low frequency response. Thus, the target variables are
the averaged switching voltage vSw and the dc output voltage for the next cycle Vout,next.
The coefficient dc is also applied here to unify the modeling of the CCM and DCM.
Depending on the states of the MOSFETs and iL, one switching cycle can be divided into
three parts:
54
1) (0 ~ d)Ts. The high-side MOSFET is on and the low-side MOSFET is off. Cout is
being charged by Vin (iL increases).
2) (d ~ dc)Ts. The high-side MOSFET is off and the low-side MOSFET is on. Cout is
being discharged (iL decreases).
3) (dc ~ 1)Ts, which only occurs in the DCM (dc < 1). The low-side MOSFET remains
off and iL already reaches zero. Vsw is equal to Vout during this period.
So the averaged switching voltage for the next cycle can be determined by combining all
the three parts:
Cv = d ( \ in - ronMiL) - (dc - d)ron,LiL + (1 - dc)Vtm. (26)
Since a dropout voltage, caused by the ESR of the inductor, exists between the switching
and output nodes, the output voltage for the next cycle can be directly calculated as:
Fout,next = ^sw D T l - ( 2 7 )
3.4. MODEL IMPLEMENTATION
To simulate the voltage droop waveform in the time-domain and the PDN
impedance in the frequency-domain, the proposed behavior model is implemented in a
circuit simulator, Keysight Advanced Design System (ADS) [20].
The schematic of this model is illustrated in Figure 7, which clearly presents all the
key components in the voltage control loop, the current control loop, and the power stage.
The analytical equations can be implemented as circuit components using the Symbolically
Defined Devices (SDD) of ADS. The “SDDnP” block indicates that this SDD component
has “n” ports. For the voltage control loop, the output voltage is sensed by a voltage-
55
controlled voltage source (VCVS) with a specific time delay (td). Then, the LPF, the
integrator, and the PI control are implemented by the SDD components © , © , and © ,
respectively. For the current control loop, the inductor current is sensed by a current-
controlled voltage source (CCVS). Sr, Sf, and TsAS are calculated by the SDD component
® . The unified equations for the CCM and DCM, (21) and (24), are fulfilled by the SDD
components © and © . At last, the SDD component © represents (26) in the power stage,
and (27) is directly executed by the circuit simulation for the next cycle Vout,next (see
Appendix for the detailed expressions of the SDD components).
By applying different simulation instances, the established circuit model can
perform the transient and ac simulations. Therefore, it provides a straight-forward
evaluation method in both the time-domain (TD) and frequency-domain (FD). This ADS
based circuit simulation is introduced as an example to implement the proposed behavior
model. Alternative implementation solutions may also be feasible based on other
simulation tools.
-nnm .
Pow er StageC u rre n t C on tro l Loop V oltage C on tro l LoopSDD6P
|O U T>L w © C C V S
C u rre n tSensor SDD6P
SD D 4PP aram eter L ist:
See T A B L E I
V C V SSD D 2PInstance:V oltageSD D 7P
I D - transientSDD3P SD D 2P Sensor FD - ac
Figure 7. Schematic o f the behavior model in the circuit simulator.
56
Table 1. VRM behavior model parameters.
Category Parameter Description
FixedParameters
K e f Internal reference voltage.
/ s Switching frequency.
f ' o u t Output filtering capacitance.L, rL Output inductor and its ESR.
U m , H > U i n . LOn-resistance of the high-side and
low-side MOSFETs.
TunedParameters
fc Cut-off frequency of the LPF.kp, k[, kfc PI control coefficients.
td Control delay.
K pRamp voltage of the slope
compensation.R, Current sensing gain.
4. CHARACTERIZATION AND VALIDATION
As discussed earlier, one of the biggest challenges in VRM modeling is the lack of
knowledge for the actual design due to the IP-protection. With the assistance of the
behavior model, the interior topology-level components are analyzed and implemented in
the circuit simulator. Table 1 summaries the model parameters, which can be classified into
two categories: the fixed and tuned parameters. The fixed parameters are accessible from
the IC’s datasheet, the PCB schematic, and other design files. The tuned parameters are
related to the internal design of the VRM and not available at the system level. In this
section, a measurement-based characterization is conducted on a power management IC
(PMIC) EVB with the target VRMs to determine the tuned parameters. Using the same
measurement setup, the behavior model with the finalized parameters is used to predict the
voltage droops under different operating conditions. By comparing the simulated and
57
measured results, the proposed modeling method is validated for both single-phase and 3-
phase VRMs.
4.1. MEASUREMENT SETUP
The measurement setup is depicted in Figure 8. The PMIC chip and its peripheral
circuit components were already assembled on the PMIC EVB. The PMIC chip contained
one single-phase and one 3-phase VRMs. Its input voltage Vin, ranging from 3.4 V to 4.2
V based on the typical battery voltage, was provided by the dc power supply (Keysight
N6705B). The VRMs could be configured by the I2C commands, such as enabling the
force-PWM mode, changing the desired output level, and setting the switching frequency.
For the single-phase VRM, the output inductor was 0.47 pH with a 31 mO ESR, the
effective output capacitance was 17 pF considering the derating effect. The output voltage
was tunable from 0.7 V to 1.0 V and the switching frequency was 2.4 MHz. For the 3-
phase VRM, the output inductor for each phase was 0.24 pH with a 18 mO ESR, the
effective output capacitance for each phase was also 17 pF. The output voltage was fixed
to 0.75 V and the switching frequency was 3.2 MHz. A custom-made slammer board was
used to draw the step transient current at each VRM’s output. The slammer boards were
directly soldered on the PMIC EVB to minimize the connection impedance. In order to
precisely control the amplitude and the slew rate of the load current, an external drive signal
V drive was provided by the signal generator (Agilent 81150A). The maximum load
current was 7 A with a minimum of 300 ns rising time. Both the output voltage and the
current were monitored by an oscilloscope (Tektronix DPO5204B). There are several
advantages in applying the time-domain measurements. First, the voltage droop in the time-
58
domain directly affects the stability of the system, so it is the most important criterion for
characterizations and validations. Also, the measurement results are more accessible by
using a less expensive instrument (i.e., an oscilloscope instead of a vector network
analyzer). In addition, the output voltage waveform in the time-domain is helpful to
determine proper initial values for the tuned parameters, which is further discussed in the
“Parameter Optimization” section.
Figure 8. Diagram and photograph of the measurement setup.
4.2. PARAMETER OPTIMIZATION
The measurement-based characterization was conducted under the nominal
operating condition. Using the single-phase VRM as an example, the input and output
voltages were 3.8 V and 0.85 V, the amplitude of the load current was 2 A, and the rising
59
time was 1000 ns. The measured output voltage and load current waveforms are shown in
Figure 9. The voltage droop in the time-domain is affected by all the model parameters.
The goal of characterization is to achieve the same voltage droop waveform by optimizing
the tuned parameters.
Conventionally, the trial-and-error method is used to test all the combinations of
parameters within the ranges of interests. However, the 7 tuned parameters are coupled
with each other, so this method is impractical and time-consuming. Therefore, a two-step
procedure, including the initial step and the fine-tune step, is introduced to efficiently
optimize the parameters.
Figure 9. Measured voltage and current waveforms for characterization.
60
1) The initial step aims to set proper initial values for the tuned parameters, so the
optimization in the next step can be expedited. Based on the characteristics of
the control loops, the measured voltage waveform in the time-domain provides
helpful information to determine the initial values, as illustrated in Figure 9.
First, the voltage droop recovery (10 ~ 20 ps) is mainly dominated by the
integrator of the voltage control loop, so ki and kdc are set based on the curvature
and final level of the recovery part, respectively. Then, a ringing behavior can
be observed after the first voltage droop (5 ~ 10 ps), which reflects the stability
of the control loop. So for the remaining parameters of the voltage control loop,
kp mainly affects the ringing frequency and td can limit the ringing amplitude.
After the initial values of the voltage control loop parameters are determined,
Ri can be selected so that the the largest voltage droop (at 5 ps) can match with
the measured result. At last, the other tuned parameters, fc and Vrp, only have
secondary effects on the voltage droop. It is suggested to set their initial values
based on the general design guidelines. For example, fc is set as two times of
the switching frequency to filter out higher frequency noise. The typical
compensating ramp slope is half of Sf [18], so the initial value of Vrp can be set
as —0.5SfTs.
2) In the fine-tune step, the ADS built-in optimization tool is used to further adjust
the parameters simultaneously based on the initial values. The goal is to
minimize the error between the measured and simulated output voltage
waveforms. The “Gradient” algorithm is chosen with maximum 200 iterations,
while other optimization algorithms are also applicable. With the proper initial
61
value set for each tuned parameter, the optimization process can quickly
converge and finalize the behavior model.
Following the same procedure, the measurement-based characterization is also
conducted for the 3-phase VRM. Since the proposed model focuses on the averaged
behavior of the VRM devices, the 3 output phases connected in parallel can be represented
by one equivalent output inductor (1/3 of the inductance and ESR in one phase) and one
equivalent capacitance (3 times of the capacitance in one phase). The optimized model
parameters for both single-phase and 3-phase VRMs are listed in Table 2.
Table 2. Characterized single-phase and 3-phase VRM models.
Category Parameter Single-Phase 3-Phase
FixedParameters
K e f 0.85 V 0.75 V/ s 2.4 MHz 3.2 MHz
C o v A 17 //F 51 f i FL 0.47 /.iH 0.08 /jH
r L 31 mO 20 m fi
'f'on.H 60 m fi 34 mf2
?’on,L 23 m fi 13 mf2
TunedParameters
f c 6.0 MHz 6.7 MHzk p 35 36
h 8.0e6 18.6e6k & c 980 775U 180 ns 288 ns
Vrp 0.22 V 0.28 VR i 0.70 0.74
62
4.3. MODEL VALIDATION
The novel method to unify the modeling of the CCM and DCM is firstly validated
using the single-phase VRM model. Then, the simulated voltage droops from the finalized
behavior models are compared with the measured results under various operating
conditions for both single-phase and 3-phase VRMs.
The input voltage was 3.8 V and output voltage was configured to be 0.85 V on the
single-phase VRM. By gradually increasing the load current from 0.05 A to 2 A, the
operation mode of this VRM changed from the DCM to the CCM at the boundary current
of 0.5 A. Figure 10 shows the simulated duty cycles for a 2 A step transient current load
under the same configuration. The dash-dotted line, which represents dc calculated based
on (21), has a turning point at 5.32 gs. When the boundary case between the DCM (dc < 1)
and the CCM (dc = 1) occurs, the corresponding averaged inductor current is 0.5 A,
which agrees with the measured boundary current. The duty cycle d, calculated based on
(25), is depicted as the solid line. When 1L = 0.05 A, d remains low at 0.08 for the DCM.
As 1L increases to 2 A and the voltage droop occurs, d is then increased up to 0.58 by the
feedback control loops. After the output voltage recovers to 0.85 V in the CCM, d is
stabilized at 0.25, which matches with the ratio of the output and input voltages considering
the additional power losses due to the MOSFET’s on-resistance and the inductor’s ESR.
Thus, the proposed unified equations are proved to model the load transition between the
CCM and DCM correctly.
63
Figure 10. Validation of the unified equations for the CCM and DCM.
Figure 11 and Figure 12 show the comparisons between the simulated and
measured voltage droop waveforms for single-phase and 3-phase VRMs, respectively.
Different operating conditions, including the output voltage level (0.7 V ~ 1.0 V for the
single-phase, fixed 0.75 V for the 3-phase), the load current amplitude (0.5 A ~ 2.0 A for
the single-phase, 4 A ~ 7 A for the 3-phase), and the load current rising time (300 ns ~
1000 ns), were investigated. It should be noted that the output voltage of the 3-phase VRM
was fixed to 0.75 V, so we instead compared the voltage droops of different rising time at
two load current amplitudes (5 A and 6 A). In addition, the voltage waveforms in Figure
11 (c), Figure 12 (b), and Figure 12 (c) are shifted by ±5 ns for better displaying purposes.
The simulated and measured results have good correlations with each other for all the
conditions. The maximum differences are within ±1.3 mV and ±1.9 mV for single-phase
64
and 3-phase VRMs, respectively. It implies that the behavior model with the optimized
parameters can accurately predict the voltage droop waveform in the time-domain.
The circuit model implemented in ADS (Figure 7) can also simulate the magnitude
and phase of the PDN impedance with the ac instance. To validate the simulated frequency-
domain result, the classic shunt-thru method was used to achieve accurate mO impedance
measurements [21]. Two semi-rigid probes were mounted at the output power and ground
pads of the single-phase VRM, the transmission coefficient S21 was measured by the vector
network analyzer (Keysight E5061B) while the VRM was in the active state. Then, the
PDN impedance was converted based on the measured S21 as: Zpdn = 25 • S21 / (1 - S21).
The frequency range in this measurement was from 50 Hz to 5 MHz. Full 2-port calibration
was applied to move the reference planes to the connectors of the semi-rigid probes. Further
port extension to the end probe tips was not required because the probe length was
electrically short in regard to the target frequency range. The simulated and measured
impedance are compared in Figure 13. Good agreements for both the magnitude and phase
can be observed up to 1 MHz, so the results validate the accuracy of the proposed model
within the desired frequency range. In addition, this model can be cascaded with the
backend PCB and package circuits, which is useful to generate the end-to-end PDN
impedance curve as illustrated in Figure 2 (b).
65
(a) Different output
voltage levels.
(b) Different load
current amplitudes.
(c) Different load
current rising times.
Measured: 2.0 A
Measured: 0.5 AMeasured: 1.0 A
Simulated: 0.5 A810Simulated: 1.0 ASimulated: 2.0 A
8 no
M easured: 300 ns M easured: 600 ns850Measured: 1000 ns
Slutted Shitted-5 ns 5 ns
Simulated: 300 nsSimulated: 600 nsSimulated: 1000 ns
815
Figure 11. Comparison of voltage droop waveforms for single-phase VRM.
66
(a) Different load
current amplitudes.
(b) Different rising
times at 5 A current.
(c) Different rising
times at 6 A current.
■Measured: 4 A 'Measured: 5 A M easured: 7 A
Simulated: 4 A Simulated: 5 A Simulated: 7 A
Time ( u s)
Shifted
Measured: 600 ns
Shitted> 720-5 ns 5 ns
-Measured: 300 ns710
Measured: 1000 ns
Simulated: 300 nsSimulated: 600 nsSimulated: 1000 ns
670
Measured: 300 ns Measured: 600 ns ■Measured: 1000 ns
Shitted Shifted5 ns
Simulated: 300 nsSimulated: 600 nsSimulated: 1000 ns
Figure 12. Comparison of voltage droop waveforms for 3-phase VRM.
67
(a) Magnitude.
(b) Phase.
Figure 13. Comparison of simulated and measured impedance for single-phase VRM.
68
In this work, the topology and modeling challenges for the VRM are discussed. To
overcome the challenges, a generic behavior model based on the current-mode topology is
developed. This behavior model includes all the key components in the voltage control
loop, the current control loop, and the power stage of an actual VRM device. A novel
method is also proposed to unify the modeling of the CCM and DCM for different load
conditions and simplify the implementation for circuit simulations. The model parameters
are optimized based on the measurement-based characterization. Good accuracy of this
model is observed by the comparisons between the simulated and measured results for both
single-phase and 3-phase VRMs. The optimized VRM model can accurately predict the
voltage droop waveforms in the time-domain and the PDN impedance in the frequency-
domain under various operating conditions. Cascading the proposed VRM model with the
PCB-level and package-level PDN models enables a combined PDN analysis, which is
much needed for modern PDN designs.
5. CONCLUSION
APPENDIX
The analytical equations of the behavior model are implemented as the SDD
components in Figure 7. In the expression of an “SDDnP” component, “vi” represents the
signal at port i (i = 1, 2, 3 ... n). The full expressions of all the SDD components are listed
below.
69
• SDD component © (SDD2P, LPF):I[1,0] = 0F[2,0] = _v2 - _viF [2 ,1] = _v2/(2 tt/ c)
• SDD component © (SDD2P, integrator):
I[1,0] = 0F[2,0] = _v2/k dc - _V] F[2,1] = _v2/k>
• SDD component © (SDD3P, PI control):
I[l> 0] — I[3> 0] — 0F[2,0] = — _v2 + min{max{_ni ■ fcp + _v3, —3}, 3}
• SDD component © (SDD6P, current feedback):
I[l) 0] = I[3,0] — I[4, 0] — 0F[2, 0] = - _ v 2 + 1 / /s - (_ui + _v3 ■ (ron.L - ?’on,H))/L F[5,0] = - _ v 5 + (_vi - _v3 ■ (ron,H + rL) - _vA)/L F[6,0] = - _ v 6 + (~_v3 ■ (ron,L + 'rL) - _Va) ) / L
• SDD component © (SDD4P, dc calculation):
I[l, 0] = I[2,0] = I[3,0] = 0_____________ _______________F[4,0] = -_ r /4+min{ a/ 2 * _v{ * (_v3 - _v2)/(_v3 ■ _v2 • f s), 1}
• SDD component © (SDD7P, duty cycle):
I[l, 0] = I[2,0] = I[3,0] = I[4,0] = I[5,0] = I[7,0] = 0 F[6,0] = —_V(, + min{max{0.5 + V^jjv-i/R i —\ / ( ° - 5 + V v p ! - V i ! R i ) 2 ~ 2/_r;7(_'t?2/ j R{ - 1}
• SDD component © (SDD6P, power stage):
70
REFERENCES
[1] M. Swaminathan, J. Kim, I. Novak, and J. P. Libous, “Power distribution networks for system-on-package: Status and challenges,” IEEE Transactions on Advanced Packaging, vol. 27, no. 2, pp. 286-300, 2004.
[2] W. D. Becker, J. Eckhardt, R. W. Frech, G. A. Katopis, E. Klink, M. F. McAllister, T. G. McNamara, P. Muench, S. R. Richter, and H. Smith, “Modeling, simulation, and measurement of mid-frequency simultaneous switching noise in computer systems,” IEEE Transactions on Components, Packaging, and Manufacturing Technology: Part B, vol. 21, no. 2, pp. 157-163, 1998.
[3] X. Zhou, P.-L. Wong, P. Xu, F. C. Lee, and A. Q. Huang, “Investigation of candidate vrm topologies for future microprocessors,” IEEE Transactions on Power Electronics, vol. 15, no. 6, pp. 1172-1182, 2000.
[4] K. Lee and A. Barber, “Modeling and analysis of multichip module power supply planes,” IEEE Transactions on Components, Packaging, and Manufacturing Technology: Part B, vol. 18, no. 4, pp. 628-639, 1995.
[5] J. Kim, L. Ren, and J. Fan, “Physics-based inductance extraction for via arrays in parallel planes for power distribution network design,” IEEE Transactions on Microwave Theory and Techniques, vol. 58, no. 9, pp. 2434-2447, 2010.
[6] K. Shringarpure, S. Pan, J. Kim, B. Achkir, B. Archambeault, J. Drewniak, and J. Fan, “Formulation and network reduction to a physics based model for analysis of the power distribution network in a production level multi-layered printed circuit board,” IEEE Transactions on Electromagnetic Compatibility, vol. 58, no. 3, pp. 849-858, 2016.
[7] S. Bai, B. Zhao, J. Cho, A. Ruehli, S. Connor, M. Cocchini, D. Becker, B. Archambeault, and J. Drewniak, “Plane-pair peec modeling for package power layer nets with inductance extraction,” in 2018 IEEE Symposium on Electromagnetic Compatibility, Signal Integrity and Power Integrity (EMC, SI & PI). IEEE, 2018, pp. 1-17.
[8] J. Kim, W. Lee, Y. Shim, J. Shim, K. Kim, J. S. Pak, and J. Kim, “Chip-package hierarchical power distribution network modeling and analysis based on a segmentation method,” IEEE Transactions on advanced packaging, vol. 33, no. 3, pp. 647-659, 2010.
71
[9] I. Erdin and M. Nakhla, “Mixed circuit/electromagnetic analysis of field coupling to high speed interconnects in inhomogenous medium,” in 1999 IEEE International Symposium on Electromagnetic Compatability. Symposium Record (Cat. No. 99CH36261), vol. 1. IEEE, 1999, pp. 446-449.
[10] M. Feliziani and F. Maradei, “Circuit-oriented fem: Solution of circuit-field coupled problems by circuit equations,” IEEE Transactions on Magnetics, vol. 38, no. 2, pp. 965-968, 2002.
[11] E.-X. Liu, X. Wei, Z. Z. Oo, E.-P. Li, and L.-W. Li, “Modeling of advanced multilayered packages with multiple vias and finite ground planes,” in 2007 IEEE Electrical Performance of Electronic Packaging. IEEE, 2007, pp. 275-278.
[12] S. Sandler, L. S. Picotest, E. Bogatin, and T. LeCroy, “Vrm modeling: A strategy to survive the collision of three worlds,” November 2018. [Online]. Available: “https://www.signalintegrityjournal.com/articles/1017-vrm-modeling-a-strategy- to-survive-the-collision-of-three-worlds” .
[13] A. Radic, Z. Lukic, A. Prodic, and R. H. de Nie, “Minimum-deviation digital controller ic for dc-dc switch-mode power supplies,” IEEE Transactions on Power Electronics, vol. 28, no. 9, pp. 4281-4298, 2012..
[14] S. Baek, P. Pun, and A. Agrawal, “Behavioral model of switching dc-dc converter for improving power delivery network design,” in 2012 IEEE 62nd Electronic Components and Technology Conference. IEEE, 2012, pp. 926-929.
[15] R. Ahmadi, D. Paschedag, and M. Ferdowsi, “Closed-loop input and output impedances of dc-dc switching converters operating in voltage and current mode control,” in IECON 2010-36th Annual Conference on IEEE Industrial Electronics Society. IEEE, 2010, pp. 2311-2316.
[16] J. Xu, S. Bai, B. Zhao, K. Nalla, M. Sapozhnikov, J. L. Drewniak, C. Hwang, and J. Fan, “A novel system-level power integrity transient analysis methodology using simplified cpm model, physics-based equivalent circuit pdn model and small signal vrm model,” in 2019 IEEE International Symposium on Electromagnetic Compatibility, Signal & Power Integrity (EMC+ SIPI). IEEE, 2019, pp. 205-210.
[17] R. Mammano, “Switching power supply topology voltage mode vs. current mode,” Elektron Journal-South African Institute of Electrical Engineers, vol. 18, no. 6, pp. 25-27, 2001.
[18] C.-C. Yang, C.-Y. Wang, and T.-H. Kuo, “Current-mode converters with adjustable-slope compensating ramp,” in APCCAS 2006-2006 IEEE Asia Pacific Conference on Circuits and Systems. IEEE, 2006, pp. 654-657.
72
[19] D. Venable, “Current mode control,” Venable Instrument Inc. Technical Paper, vol. 5, 2009. [Online]. Available: “https://venable.biz/uploads/files/05-Technical- Paper-Current-Mode-Control.pdf’.
[20] K.-W. Yeom, Microwave Circuit Design: A Practical Approach Using ADS. Prentice Hall Press, 2015.
[21] Keysight. (2014) Evaluating dc-dc converters and pdn with the e5061b lf-rf network analyzer. [Online]. Available: http://literature.cdn.keysight.com/litweb/ pdf/5990-5902EN.pdf.
73
III. ACCURATE RECTIFIER CHARACTERIZATION AND IMPROVED MODELING OF CONSTANT POWER LOAD WIRELESS POWER TRANSFER
SYSTEMS
ABSTRACT
Wireless power transfer (WPT) is an emerging technology deployed for a wide
range of applications. Various design challenges need to be addressed at both circuit and
system levels. However, the conventional modeling methods either require impractical
assumptions or only focus on the coil-to-coil part. This article presents a system-level
model for WPT applications, including all the critical interior blocks, such as power
amplifier, transmitter coil, receiver coil, matching networks, and rectifier. An accurate
characterization method is proposed to obtain the rectifier impedance as a realistic load
condition. Then, through frequency harmonic analysis, the power capabilities along the
power flow path are analytically derived for both the fundamental and higher order
harmonic components. The system efficiency and the power loss at each block can be
accurately estimated. As a result, this model can not only optimize the system performance,
but also help improve the thermal design for WPT products. With the assistance of this
improved model, a practical design methodology is introduced to optimize the system
parameters. An experimental prototype is built to validate the proposed model and the
design methodology. Good correlations are observed between the calculations and
experiments in both the time-domain and frequency-domain results.
74
1. INTRODUCTION
Wireless power transfer (WPT) is an emerging technology that brings great
convenience toc harging a large variety of electronic devices. It has been widely deployed
for domestic and industrial applications, including cell phones, home electronics, medical
implants, and electric vehicles [1 ]—[4]. These applications usually have different
requirements on power level, efficiency, physical size, and operating frequency. WPT
standards, such as Qi from Wireless Power Consortium [5] and J2954 from the Society of
Automotive Engineers [6], have been proposed to improve the interoperability and the
safety protection of WPT applications.
Figure 1. Typical WPT system with the SS topology.
In addition to the standards and regulations, it is crucial to minimize the power loss
and improve the efficiency of WPT systems, since the dissipated heat of devices may cause
potential safety issues and throttle the charging performance. There are three major
challenges in designing a power efficient WPT system. First, it remains an unresolved
question to achieve the optimal efficiency at the system level. As shown in Figure 1, a
75
typical WPT system is formed by multiple blocks, including the power amplifier, the
magnetic coils, the matching networks, and the rectifier. It is worth noting that the optimal
configuration of a single block may not correspond to the peak efficiency of the entire
system. Second, the power loss of each block in a WPT system needs to be estimated
accurately. Because of the increasing power levels, the thermal design becomes more
important for the electronic products. An effective thermal design needs to account for not
only the end-to-end total loss, but also the dissipated power at each block along the power
flow path. Third, there are no practical methodologies to simultaneously optimize multiple
system parameters, such as the coil matching capacitance, the operating frequency, and the
input voltage level. The conventional trial-and-error method requires extensive
measurements on a prototype system, which is time-consuming and expensive. These
challenges limit the performance for a wide range of WPT applications, including the cell
phones and electric vehicles [3]. To overcome these challenges, an accurate system-level
model is desired to conduct the global optimization and loss estimation for all the blocks
in a WPT system.
Over the past few decades, various studies have been carried out to investigate WPT
designs at both coil and system levels [7]-[12]. Among these studies, a magnetic coil
design methodology for the WPT system was presented in [7]. In addition to magnetic
coils, matching networks and other system parameters are also critical to maximize
efficiency. In [8] and [9], the power transfer capability was analyzed under different
matching topologies for the coupled coils. However, the output load of the coils was
assumed to be purely resistive, which is impractical because the nonlinear rectifier device
may contribute reactance components to the load in real WPT systems. The effects of
76
rectifier input impedance were discussed in [10] and [11]. However, the authors in [10]
only focused on the resonant condition and ignored the case when nonzero line impedance
is present at the ac side of the rectifier. In addition, the rectifier impedance was obtained
from PSPICE simulations, but the PSPICE model may be unavailable for many devices
because of the lack of support or confidential concerns. In [11], an analytical model of the
rectifier impedance was established for the positive half-cycles, but it neglected the
nonlinearity caused by the line-commutation process. Zhang et al. presented analytical
expressions of coil efficiency with both frequency and voltage tuning [12]. The derivations
were based on the first harmonic approximation (FHA), which becomes inaccurate when
the operation condition is not at the resonant frequency [13]. Other conditions from the
existing WPT products also make FHA inaccurate, such as the low-voltage operation of
the cell phone wireless charger, the low power charging mode that causes the loaded quality
factor decreasing [14], and the frequency splitting and bifurcation phenomena due to high
coupling coefficient [15].
This article aims to provide a comprehensive discussion on the system-level
modeling for kilohertz WPT applications. The limitations of the existing models are
investigated based on circuit-level simulations in Section 2. To overcome these issues, an
accurate characterization method for the rectifier is proposed in Section 3. It provides a
feasible solution to obtain the rectifier impedance, which is included as a realistic load
condition for the improved WPT system model in Section 4. Using this model, a design
methodology is introduced to determine the optimized parameters for a WPT system. Then,
an experimental prototype is implemented to validate the proposed model and the design
methodology in Section 5. Finally, Section 6 concludes this article.
77
2. TOPOLOGY AND PREINVESTIGATION
A typical WPT system (see Figure 1) consists of four blocks: 1) the power
amplifier, to convert dc input voltage Vin and current Iin to ac power, which is commonly
implemented by half-bridge or full-bridge topology; 2) the transmitter (TX) coil with self
inductance Ltx and matching capacitance Ct x ; 3) the receiver (RX) coil with self
inductance Lrx and matching capacitance Crx; and 4) the rectifier, to inversely regulate ac
power to dc output voltage Vout.
In this article, the series-series (SS) matching topology is used. Other matching
topology, such as parallel-series, series-parallel, and parallel-parallel, can be derived in a
similar manner [8], [9]. For the remaining parameters in Figure 1, Rtx and Rrx are the
equivalent series resistances (ESRs) of the coils and matching capacitors. k is the coupling
coefficient of the coupled magnetic coils. Then, the mutual inductance can be expressed as
M = k^L txLrx. Cload is the smoothing capacitor at the rectifier’s output. It is a common
practice for WPT applications to add sufficiently large Cload to stabilize the output voltage,
so the output voltage ripple can be neglected. Zrect is the input impedance seen at the
rectifier input port. Figure 1 also illustrates the power flow of a WPT system, including TX
input power Pin, power amplifier output Ppa, transferred power to the RX side Prx, rectifier
input power Prect, and RX output power Pout. The rated output power is typically specified
as a requirement in the early design stage, and the output voltage is constant in WPT
systems [16]-[18]. For the design purpose of efficiency optimization under the maximum
output power, this article focuses on the constant power load (CPL) condition with fixed
output voltage.
78
To investigate the limitations of assumptions used in the existing methods, a circuit
model is built in an RF simulator, Keysight Advanced Design Systems, based on the WPT
system with the SS topology. As some WPT models only focused on the coil-to-coil part,
the system efficiency is compared with the coil-only efficiency to show the significance of
other blocks in the system. Then, the inaccuracy of FHA is investigated by the comparison
between the time-domain and frequency-domain results. Finally, by applying different load
conditions to the RX coil’s output, the simulations demonstrate that inaccurate estimation
of the efficiency can be caused by the pure resistive load assumption. Therefore, it is
important to model the WPT system with a realistic load impedance.
TXSide R X Side Load (I): Rectifier Load
R.„.=1.5xl0 f - 0.13si c k=0.64— W v R li =
20 uF 8.67 QC^=400 nF ( ’ =300 nF
SourceW V
f = 120 L,,=8.6 FI I L,,= 12.5 FI I180 kHz
R,,=70 mL2 R ,,=270 m ilLoad (II): Resistor Load
Instance (1): TransientR n _Instance (2): Harmonic Balance
7.60 £2
Figure 2. Diagram of the circuit model for preinvestigation.
79
The configuration of this circuit model is depicted in Figure 2. The value of each
element is chosen based on a Qi-standard cell phone WPT system. This model can be
simulated in both time and frequency domains by different simulation instances. On the
TX side, the power amplifier is simplified to a trapezoidal-wave voltage source. The
amplitude of this voltage source is between —Vin and Vin; the operating frequency is f. A
frequency-dependent equivalent resistor Rsrc is used to account for the power loss caused
by the power amplifier. The function of Rsrc is determined by fitting the simulated power
loss to the measured power loss of an existing power amplifier. Even though the equivalent
resistor simplification cannot accurately model the loss of the power amplifier, it is
sufficient to preinvestigate the limitations of the existing methods. A more rigorous
modeling method is discussed in Section 4. On the RX side, the output power is fixed at 8
W. The CPL condition is satisfied by tuning Vin and f. Two load configurations are applied:
(I) a full-bridge rectifier (D1-D4) with the load resistor RL1; and (II) a pure resistor load
RL2 directly connected to the RX matching network. To ensure similar output power levels
for the two load configurations, RL2 is converted from RL1 based on the approximated
effective resistance equation [14]:
(1)
where Vfw = 0.34 V is the forward voltage of the rectification diode, and Vout = 8.33 V is
the rectifier output voltage.
80
(a) Comparison of
system and coil-to-coil
efficiencies.
Figure 3. Simulation results of the circuit model.
Coi
l Eff
icie
ncy
(%)
81
2.1. INVESTIGATION OF SYSTEM AND COIL EFFICIENCY
Figure 3 (a) compared the system and coil efficiencies at different frequencies using
load configuration (I). The system and coil efficiencies, n sys and n coil, were calculated as
isys —
coil =
• 100% ,
■ 100% ,
(2)
(3)
where Pin and Pout are the input and output powers, respectively, and Ppa and Prect are the
power amplifier output and rectifier input powers, respectively.
Note that the peak system efficiency is achieved at a lower frequency compared
with the peak coil efficiency. This implies that the TX and RX coil blocks are not the only
factors determining the efficiency for a WPT system. Other blocks also have significant
contributions; one important loss term related to frequency is the MOSFET switching loss
of the power amplifier [19]. As the frequency increases from 120 to 180 kHz, the simulated
loss of the power amplifier reaches up to 583 mW. In addition, the design of coil’s matching
network also affects the current flowing through all the blocks, which could affect the loss
of the power amplifier and the rectifier. In the simulations and optimizations for real-world
WPT applications, a system-level model is preferred over a coil-to-coil only model.
2.2. INVESTIGATION OF HARMONIC FREQUENCY
The simulated system efficiencies in the time and frequency domains, using the
load configuration (I), were compared in Figure 3 (b). For the time-domain simulation, the
steady-state voltage and current waveforms were directly multiplied in a unit time step to
82
calculate the input and output powers. The system efficiency in the time domain was
obtained based on (2). For the frequency-domain simulation, with the dc output power
fixed to 8 W, the input power carried by each frequency component was calculated
individually by the magnitudes and phase difference of the simulated voltage and current.
Then, the system efficiencies in the frequency domain were determined using the input
power of the fundamental frequency only (“FD 1st only”), the input power including up to
the third-order harmonics (“FD up to 3rd”), and the input power including up to the fifth-
order harmonics (“FD up to 5th”).
As shown in Figure 3 (b), only at the crossing point, the efficiency calculated by
the fundamental frequency component can match with the time-domain efficiency. It is
because the input power of higher order harmonics is zero due to the 90o phase angle
between the voltage and current, and only the fundamental frequency component carries
the input power. With the increasing offset to this condition, the difference of the
efficiencies using the fundamental frequency component and using the time-domain
component becomes larger. It implies that FHA is not applicable, and higher order
harmonic components should not be neglected. The inaccuracy of FHA is mainly due to
two reasons. First, the voltage source of a real WPT system is not sinusoidal. Second, the
nonlinearity of the rectifiers is not taken into consideration in FHA [20]. Therefore, a valid
WPT model should have the capability to characterize the power distributed at both
fundamental frequency and higher order harmonics. The waveform in a WPT system
typically has a 50% duty cycle, so the power carried by even harmonics is negligible. For
all practical purposes, the fundamental, third-order, and fifth-order harmonic components
are the most critical terms in a WPT model.
83
2.3. INVESTIGATION OF RECTIFIER IMPEDANCE
The system efficiencies of two load configurations were compared in Figure 3 (c).
Since the circuit of the load configuration (II) has no rectifier, to maintain the validity of
this comparison, the output power in this case was estimated using an artificial full-bridge
rectifier with a forward voltage Vfw of the rectification diode
p L oad (II) _ -‘ out = PL2
Vout + 2 V f w(4)
where PL2 is the simulated power at RL2, and Vfw and Vout are defined the same as (1).
Then, the system efficiency can be calculated based on (2).
In load configuration (I), the simulated rectifier reactance varies from 0 to 1.34 Q
at different frequencies. The nonzero reactance at the RX side could affect the reflected
impedance at the TX side and change the coil efficiency. With the absence o f this rectifier
reactance, in load configuration (II), the coil efficiency has less variation at different
frequencies. As a result, the pure resistor load model gives a misleading estimation for the
system efficiency and the optimal operation point. Therefore, a realistic load configuration
is important to accurately model the efficiency in WPT systems. Both real and imaginary
parts of the load impedance should be included in the model. It is also desirable to develop
a practical characterization method for the provided rectifier diodes or ICs, in order to
obtain the accurate impedance under various conditions.
84
3. ACCURATE RECTIFIER CHARACTERIZATION
The rectifier is a common nonlinear device in ac-dc power systems. The
nonlinearity of the rectifier is mainly caused by the parasitic of the rectification diode [21]
and the line-commutation process [22]. Figure 4 depicts the circuit model of a full-wave
diode rectifier. The source voltage is denoted as vs. At the dc output port of the rectifier,
there are output voltage vdc and output current idc. The dc component of idc is Id. The load
resistance and the smoothing capacitor are Rload and Cload, respectively. At the input port
of the rectifier, there are ac voltage vac and ac current iac. The ac-side line impedance is
defined as Zline = Rline + jXline. The line resistance Rline will cause the voltage drop from
the source to the rectifier input. Since its value is determined by the ac resistance of the
coils and the ESR of the matching capacitors, which is typically tens of milliohm, Rline
can be neglected as a common practice. Xline is calculated by the line impedance of the RX
side and the reflected impedance from the TX side as
(5)
where Ltx and Lrx are the TX and RX coil inductances, respectively, Ctx and Crx are the
TX and RX matching capacitances, respectively, M is the mutual inductance between two
coils, and m is the angular operating frequency. Depending on the coil inductance,
matching network, and operating frequency, nonnegligible Xline may present at the input
of the rectifier.
85
Based on the small-signal analysis [22]-[25], the source voltage vs consists of a
sinusoidal wave v 1 and a small-signal perturbation Avp: vs = v 1 + Avp. The amplitude of
Avp is much smaller than that of vx: jv j = V^ |Avp| = Vp with Vp « Vx. The rectifier
impedance Zrect can be expressed as follows (see the Appendix for the full derivation):
(6)
where S and AS are the nonlinear switching functions; S accounts for the line-
commutation process and the parasitic of the rectification diode [22], and AS accounts for
the change of nonlinearity caused by the input perturbation [24]. v 1 and vp can be further
expanded into the complex forms at the perturbation frequency with the amplitudes V and
Vp, respectively. The nonlinearity of perturbation AS is proportional to Avp with the same
amplitude, so Vp is canceled. As a result, Zrect is a function of the operating frequency, the
input voltage amplitude Vx, the dc load current Id, the load impedance Rload, and the input
line impedance Zline.
86
With the focus on the main nonlinear behaviors caused by the line-commutation
process and the parasitic of the rectification diode (S ^ 0, AS = 0), (6) can be simplified
to (7). In addition, the line resistance Rline can be neglected as a common practice, so
Zline = jXli
(7)
It implies that the rectifier impedance is mainly affected by Xline and load impedance Rload.
With the CPL condition applied to the WPT systems, Xline becomes an effective criterion
to characterize the rectifier impedance.
Rrect SimulatedRrect MeasuredXrect SimulatedXrect Measured
'v 'll
-0.52 30 40 150 70
Frequency (kHz)
Figure 5. Comparison of simulated and measured rectifier impedance.
87
As discussed in [22], a standard sign function is used to represent the line-
commutation process of the nonlinear switching function S. However, the analytical model
described in this article is based on the assumption that no input line impedance is present
at the ac side of the rectifier. It is not applicable to the kilohertz WPT system because Xline
is commonly nonnegligible when the system is operating in an off-resonance condition.
Another method to obtain the rectifier impedance is from the circuit simulation, where
users need to correctly set all the parameters for the target device under simulation. Some
electrical parameters, such as forward voltage drop (Vfw) and reverse break down voltage
(Bv), are accessible from the datasheet. However, some physical parameters, such as
junction grading coefficient (m), emission coefficient (N), saturation current (Is), parasitic
resistance ( Rs), zero-bias junction capacitance ( Cj0), and transit time ( Tt ), are either
unavailable or require complicated calculations based on the figures in the datasheet. To
investigate the accuracy of the circuit simulation for the rectifier impedance, a full-bridge
rectifier using four Schottky diodes (SBR3U40P1) was simulated. In spite of the specified
parameters, the unavailable parameters, including m , N , and Tt , were set using the
recommended values from [26]. Other parameters, such as Is, Rs, and Cj0, were calculated
or extracted based on the typical characteristics and capacitance figures from the datasheet
as m = 0.333, N = 2, Is = 1.8e — 3 A, Rs = 34 mfi, Cj0 = 80 pF, Tt = 0 ns, Bv = 40 V.
The operating frequency varied from 120 to 170 kHz. The coil inductance, matching
networks, and Xline were maintained the same for both simulation and measurement setups
at each frequency. The comparison between the measured and simulated results is depicted
in Figure 5, where only the fundamental frequency is considered. The rectifier impedance
(Zrect) is compared by its real and imaginary parts (Rrect and Xrect), respectively:
88
^ r e c t — - ^ r e c t H " j X r e c t - ( 8 )
Large discrepancies are shown between the simulated and measured rectifier impedances.
There are mainly two reasons for the inaccurate results. First, some parameters are not
specified in the datasheet; setting recommended values based on user’s experiences or
extracting values from the figures will introduce errors. Second, the testing condition in
the datasheet may not be the same as the operating condition in measurement. Thus, the
parameters may change in a real WPT system. In sum, the analytical equation and circuit
simulation are not capable of obtaining the rectifier impedance accurately.
In this article, a measurement-based characterization method is proposed for the
rectifier. It provides the rectifier impedance and efficiency under the real operating
condition. The characterization setup is illustrated in Figure 6. In practice, the load of the
rectifier may consist of a dc-dc voltage regulator charging a battery. The dc-dc voltage
regulator, such as buck or buck-boost converter, can be modeled as an equivalent resistor
[27], [28]. Therefore, a resistive load Rload is used to represent a subcircuit feeding the
actual physical electric load of the rectifier [29]. A differential voltage probe and a current
probe are applied to the input port of the rectifier. The ac input voltage and current
waveforms are measured by the oscilloscope in the time domain. Then, all stored data are
processed by fast Fourier transformation (FFT) to provide the impedance in the frequency
domain. There are several advantages in applying the time-domain measurements. First,
the phase information is accessible by using a less expensive instrument (i.e., an
oscilloscope instead of a vector network analyzer). Also, with the relatively high signal-to-
noise ratio signals operating in the kilohertz frequency range, the phase information could
be obtained without the down mixing method. In addition, the measurement time is
89
reduced, since a single-shot waveform in the time domain contains the complete
information in the frequency range of interest (both fundamental and higher order harmonic
components).
Table 1. Coil parameters for characterization and validation setups.
CoilParameter
Generic Coil Module for Characterization
WPT System Coil for Validation
Lix 7.90 /aH 8.58 i i HLrx 7.90 12.52 /aHOx 400 nF 400 nF
Cn 100, 200, 3(K), 400 nF 200 nFk 0.58 0.63
90
The tunable LC resonant circuit is implemented to vary Xline. It is preferable to use
a generic coil-to-coil module because of three main reasons. First, the line impedance Xline
can be easily changed by the matching capacitance and the topology. Second, the scenario
for characterization is close to the real WPT system, so it can achieve good coverage of the
interested parameter range. Third, there is no additional hardware or cost. The parameters
of the generic coil used in the rectifier characterization are described in Table 1. The line
impedance Xline is tuned by varying the operating frequency and RX matching capacitance
Crx. The characterization results were validated by a real WPT system with a different set
of coils.
In Figure 7, the real and imaginary parts of the rectifier impedance (Rrect and Xrect)
are characterized separately. The part number of rectification diode is DIODES
SBR3U40P1. Vout = 8.3 V, Pout = 8 W, Zload = 8.67 H, operating frequency is 120 kHz
~ 180 kHz. The circular dots are the characterization points using the generic coil module.
The solid line is the fitted impedance curve based on the characterization points. The
triangle dots are the measured rectifier impedance of a real WPT system. Both Rrect and
Xrect at the fundamental frequency have strong correlation with Xline [see Figure 7 (a) and
(d)]. Then, the impedance at the third- and fifth-order harmonics can be further predicted
by Rrect and Xrect at the fundamental frequency [see Figure 7 (b), (c), (e), and (f)]. Thus,
Zrect can be accurately obtained in terms of its most critical frequency components for a
WPT model.
Xrec
t 3rd
(£2)
Rrec
t 5th
(£2)
Rrec
t 1st
(£2)
91
Figure 7. Rectifier impedance characterization and validation results.
92
The rectifier efficiency (nrect) is mainly affected by the forward voltage drop of the
rectification diode, the output voltage of the rectifier, and the power level. This article
focuses on the CPL WPT system, where the output power and voltage are fixed. The
characterization is also conducted on a provided rectifier with the predetermined forward
resistance property. So, n rect is expected to be constant under different operation
conditions. The measured n rect in both characterization and validation setups was between
93.6% and 94.0%. For simplicity, the averaged value 93.8% was chosen as the
characterized rectifier efficiency.
Currently, the characterization curves are fitted using polynomial and rational
functions, which are sufficient for kilohertz WPT systems. However, certain discrepancies
were observed at higher frequency harmonics [see Figure 7(c) and (f)]. It implied that more
complicated fitting or data training techniques should be applied for the WPT system with
higher operating frequency (megahertz or above). This part of work is beyond the scope of
this article and will be discussed in future studies.
4. IMPROVED MODELING AND DESIGN METHODOLOGY
In this section, an improved WPT system model is demonstrated with the
characterized rectifier impedance. This model is established based on the premise that all
the blocks of a target WPT system are provided. Then, the coil parameters, the impedance
and the efficiency of the rectifier, and the parasitic effects of the power amplifier can be
obtained in advance (see Table 2: “Known Parameters”). The goal of this method is to
optimize the matching network, input voltage, and operating frequency (see Table 2:
93
“Design Parameters”), in order to obtain 425 the peak system efficiency and accurate
power loss of each block 426 (see Table 2: “Target Outputs”). A practical design
methodology is introduced to optimize these system parameters.
Table 2. Categories of the WPT system model parameters.
Category Parameter Description
K now n
P aram eters
L t x , L pc, R t x , R r x - , k
C oil in d u c ta n ce and ac re s is tan ces , co u p lin g co effic ien t
recti ^frecti O r e e l R ec tif ie r im p e d a n ce and effic iency
i f 1 1 dead ? > R o n
P o w er am p lifie r fa ll/d ead tim es, M O S F E T b o d y -d io d e fo rw ard
vo ltage and o n -re s is tan c e
D esign
P aram eters
V m T X dc in p u t vo ltage
u — 2 i r f O p era tin g freq u en cy
O x , a *C a p a c ita n ce o f th e T X and R X
m a tch in g n e tw o rk s
T arget
O u tpu tsR m i Tpai P rx; -frect P o w ers at the b lo c k in te rfaces
V s y sS y stem effic iency
4.1. IMPROVED MODELING OF THE WPT SYSTEM
A typical WPT system is formulated in Figure 8. As discussed in Section 2, the
power amplifier at the TX side can be simplified to a trapezoidal-wave voltage source.
Since the rising/falling time of the input voltage waveform (typically tens of ns) is
relatively small compared with the switching period (several ps for a kHz WPT system),
the input source can be modeled as a square-wave with additional losses due to the
transition edges. The original dc input voltage Vin then becomes the amplitude of this
square-wave. Rs is related to the on-resistance Ron of the MOSFET in power amplifier. It
94
leads to the voltage drop along the power flow path and affects the back-end blocks. For a
half-bridge power amplifier, Rs = Ron. For a full-bridge power amplifier, R s = 2Ron. The
rectifier impedance at the RX side is Zrect, which is included in this model as a realistic
load condition. The characterization method for Z rect and rectifier efficiency n rect are
introduced in Section 3.
Figure 8. Formulation for the improved modeling of the WPT system.
Based on the standard mutual inductance coupling transfer model, the RX side
circuits can be represented as a reflected impedance Zrx ref at the TX side
(9)
95
Combining (8) and (9), Zrx_ref can be represented as:
(10)
where Rrx_ref and Xrx_ref are the real and imaginary parts of the reflected impedance
respectively.
Note that (10) is applicable to one single frequency. In order to take multiple
harmonic frequencies into consideration, a superscript (n) is introduced into each notation
to denote the index of the harmonic frequency. For example, the variables X(1), X(3), and
X(5) are the fundamental, third order, and fifth order frequency harmonics, respectively.
Then, Rrx_ref and Xrx_ref can be represented as
(11)
where u>0 is the fundamental component of the operating frequency.
96
The total load impedance o f the power amplifier at the TX side is denoted as Zpa.
Similar as the reflected impedance, Zpa can be represented with the real and imaginary
terms
where
(12)
As the input source is modeled as a square wave. The power losses related to the
transition edges are calculated separately. Using the Fourier transform, the input source of
a square wave can be divided into multiple sinusoidal harmonics, with the amplitude of
each harmonic being
y i n , =
v s ?7r n(13)
where C is a constant determined by the power amplifier’s topology. For a half-bridge
power amplifier, C = 2 as the switching voltage changes from 0 to Vin. For a full-bridge
power amplifier, C = 4 as the switching voltage changes from — Vin to Vin. With (13), the
input power distributed on each harmonic component can be calculated as
(14)
97
The time-domain input power is the sum of all the frequency harmonics
OOp _\ ' p(n)-Tin,cond — Mn.cond
71=1
= ( c r p 2 ”2tt2
n — 1
(15)
where and are shown in (12). The full expansion of Pin,cond is straightforward
and will be omitted here.
Note that Pin,cond only considers the conduction loss of the power amplifier. To
model the total input power of the WPT system, the switching loss Ploss, sw and the dead
time loss Ploss,dead of the power amplifier should also be included [30]
-Pin = -Pin,cond "T -Ploss,sw T -Ploss,dead- (16)
Dead Time ConductionSwitchingLossLoss Loss
in /
| T R ory‘
? SWGateV.sw Driver
V d - t TimeL e a d
swTime
Sine wave, f
Figure 9. Time-domain voltage and current waveforms at the switching node.
98
The switching loss and the dead time loss occur at the MOSFET on-off transitions.
For a zero-voltage-switched power amplifier, Ploss,sw is mainly contributed by the turn
OFF switching losses of the high-side and low-side MOSFETs. And Ploss,dead is caused by
the current flowing through the body diodes during the excessive dead time after each
transition. As shown in Figure 9, Ploss,sw and Ploss,dead for a half-bridge power amplifier
(one pair of high-side and low-side MOSFETs) can be approximated using simple time-
domain figure calculations. For a full-bridge power amplifier, Ploss,sw and Ploss,dead need
to be doubled compared with the half-bridge power amplifier
Mn^pa^f/, half-bridge power amplifier 2Vin/pat f / , full-bridge power amplifier
2Kifp;Tdead.A half-bridge power amplifier 4Vd/paidead/5 full-bridge power amplifier
(17)
(18)
where t f is the falling time of the on-off transition. t dead is the dead time after the
rising/falling transitions. Vd is the forward voltage of the MOSFET body-diode. Vin and f
are the design parameters: input voltage and operating frequency. Ipa is the switching node
current Isw at the on-off transition edge
4pa — - sw( — ^on—K)ff)- (19)
Isw is the composition of a sine wave with the operating frequency f and its higher order
harmonics. The amplitude of each frequency component is denoted as 1 ^ , which can be
calculated as in [14]
,<„> = F („)CosH>W C V .co s 'P Wm Ks D(n) 7r D<™)-Rpa 71 n R p a J
(20)
99
(21)
where ^ (n) = tan -1 (x p a /R p ^ ) is the phase difference between voltage and current at the
switching node.
Suppose that the MOSFET on-off transition occurs at time t on^ off = 1 / (2 /) . From
(19)-(21), Ipa is calculated using the following formula:
(22)
Integrate Ipa into (17) and (18) to determine the Ploss,sw and Ploss,dead, respectively. Finally,
based on (16), the total input power of the WPT system Pin can be obtained.
The power flowing through the other blocks, such as the power amplifier, WPT
coils, and rectifier, can be derived in a similar manner as
(23)
(24)
100
(25)
The output power is a result of multiplying the rectifier input power by the characterized
rectifier efficiency
-^out = ^rect-frect- (26)
The proposed WPT system model can accurately derive the power at each block by
using the known and design parameters in Table 2. So, the power loss and system efficiency
can be calculated by the two corresponding power values. For instance, the power loss on
the TX coil and TX matching network is (Ppa — Prx), the power loss on the RX coil and
RX matching network is (Prx — Prect) , and the system and coil efficiencies can be
calculated based on (2) and (3) respectively. Therefore, this model is capable of conducting
system-level optimization by tuning the design parameters analytically, instead of building
a complete prototype and conducting massive measurements.
4.2. PRACTICAL DESIGN METHODOLOGY
One of the challenges in designing WPT systems is the lack of practical design
methodologies. There are multiple parameters, such as the input voltage, the TX/RX coil
matching capacitance, and the operating frequency, involved in determining the system
efficiency. These parameters are coupled with each other, so it is unfeasible to tune them
one by one. Conventionally, the trial-and-error method is used to conduct measurements
101
under all the combinations of parameters within the ranges of interests, which is time
consuming and costly. With the help of the improved WPT model, a novel design
methodology is introduced to overcome these challenges.
Figure 10. Flowchart of the design methodology.
The entire design process is divided into three phases: the pre-characterization
phase, the initial phase, and the fine-tune phase. This design flow is shown in Figure 10.
1) The pre-characterization phase aims to obtain the “known parameters” (see
Table 2) of the provided blocks, including the power amplifier, the coil module, and the
rectifier. These parameters are accessible from the device’s datasheet, oscilloscope, or
impedance analyzer measurements, and the characterization method discussed in Section
3. The pre-characterized data can be collected into the lookup tables.
2) After the characterization on provided blocks, it is critical to set proper initial
values for the design parameters. In the initial phase, the WPT model can be simplified by
neglecting the switching and dead time losses of the power amplifier, and the higher order
frequency harmonics. Then, the simplified system efficiency becomes
102
(27)
(1) (1)Note that n rect, Rs , R-x , and Rtx are constants unrelated to the design parameters.
Therefore, a figure of merit (FOM) is defined to find the initial values for the design
parameters as
(28)
(1) (1)where A = R-x) and B = Rs + R-x) are both pre-determined constants. In this simplified
condition, the goal is to maximize the FOM at the fundamental frequency. Based on (5)
and (11), the FOM is determined by Crx, Ctx , and w. Other design parameters, such as Vin,
are mainly set to provide sufficient power capability for the WPT system.
3) Finally, the full WPT model is included in the fine-tune phase. All the design
parameters are tuned in fine steps based on the initial values. So, the optimizations can
quickly converge to the actual optimal efficiency point.
5. EXPERIMENT AND VALIDATION
An experimental prototype was implemented to validate the proposed model. It was
designed based on a real WPT system for cell phone products. In addition, the design
methodology to optimize the system parameters is also validated using this prototype.
103
OscilloscopePower Supply
Signal Generator
Current Probe Diff Probe
Coil Module
Figure 11. Photograph of the experimental prototype.
The experimental setup is depicted in Figure 11. The coil module consists of one
TX coil and one RX coil. Two coils are aligned with each other by the center point and
have 3-mm spacing in the vertical distance. The TX coil is built based on the standard A11
type from the Qi specification [5]. The RX coil is routed by 15-turn solid copper wires, and
the physical dimensions are 38 mm x 32 mm. Both the TX and RX coils are assembled
with magnetic ferrites. The matching capacitance is 400 nF for the TX coil and 200 nF for
the RX coil. The detailed electrical specifications of this coil module are listed in Table 1
(WPT system coil for validation). On the TX printed circuit board (PCB) board, a full-
bridge power amplifier is implemented by four N-channel MOSFET s (AON7544) and two
gate drivers (NCP81151). The dc voltage for the TX board is provided by a dual-output
power supply (Agilent E3648A). Since the proposed modeling method focuses on the
power flow path of the WPT system, the experiment is also designed to remove the impact
104
of other nonpower-path circuits. Therefore, one output of the power supply is dedicated to
the power flow path. The non-power-path devices, such as the gate driver, are powered
separately by the other output. The gate drivers are controlled by two inverted channels
from a signal generator (Agilent 81150A), where two square waves with the same
frequency and 180o out phase are generated. On the RX PCB board, a full-wave rectifier
is implemented by four Schottky diodes (SBR3U40P1) and two 10 pF smoothing
capacitors. The dc output is connected to an electrical load (KIKUSUI PLZ164WL), which
operates in the 8.67 Q constant-resistance mode. The RX output voltage is fixed to 8.33 V,
and the rated output power is 8 W. The system efficiency of this experimental setup is
calculated based on the dc powers measured at the input and output sides (Pin/P out). The
intermediate ac powers, such as the power amplifier output power (Ppa) and the rectifier
input power ( Prect), are measured in the time domain by an oscilloscope (Agilent
MSO8104A) with differential and current clamp probes.
5.1. VALIDATION ON CHARACTERIZATION AND MODELING
In order to validate both the rectifier characterization method and the improved
WPT model, three types of results have been compared: 1) the direct measured power and
system efficiency from this experimental setup: “Exp Direct”; 2) the calculated result with
measured rectifier impedance: “Cal w/Measured”; it validated this improved WPT model;
and 3) the calculated result with characterized rectifier impedance: “Cal w/ Characterized”;
by using the fitted rectifier impedance curve (the solid line in Figure 7), it validated the
accuracy of this rectifier characterization method.
105
(a) Total input power.
(b) Total output power.
(c) System Efficiency.
Figure 12. Comparison of dc power and efficiency.
106
The calculations, based on (16), (26), and (2), and the experiment results of input
power Pin, output power PouU and system efficiency n sys versus the operating frequency
are plotted in Figure 12. The rated output power is 8 W with less than 0.1 W variation in
actual measurement, and the operating frequency varies from 120 to 170 kHz. The three
types of results have good correlation with each other. The highest n sys is achieved at 150
kHz, with the input voltage Vin = 8.52 V. The improved WPT model does not require any
impractical assumptions used in the conventional modeling methods and can find the peak
efficiency point across a wide frequency range.
Since this model is built on the basis of frequency harmonic analysis, it is important
to validate the accuracy of the calculated power distributed on each frequency component.
The comparison of intermediate ac powers, including Ppa and Prect, is shown in Figure 13.
For experiment, the time-domain result (“Exp TD”) is obtained by multiplying the
measured voltage and current directly, and the fundamental frequency result (“Exp 1st”) is
extracted from the FFT processed time-domain waveforms. For calculation, Ppa and Prect
distributed on each frequency component are calculated based on (23) and (25). The time-
domain result (“Cal TD”) includes the three most significant components: the fundamental,
third-order, and fifth-order harmonics. The fundamental frequency result (“Cal 1st”) only
considers the fundamental component. The good agreement between experiment and
calculation validates the improved model in both time and frequency domains.
Additionally, it shows that this model can not only estimate the end-to-end system
efficiency, but also provide the accurate power loss of each block in a WPT system. As a
result, it provides more flexibility to set different loss thresholds for the critical blocks and
107
improve the thermal design. The configuration with the maximum FOM achieves the best
system efficiency.
(a) Power amplifier
output power Ppa.
Exp Direct TDCal w/ Measured TDCal w/ Characterized TDExp Direct 1stCal w/ Measured 1 st
■■ Cal w/ Characterized 1st
9 .4
V 1- ■ ■-•A
Frequency (kHz)
(b) Rectifier input
power Prect.
o Exp Direct TDCal w/ Measured TDCal w/ Characterized TDExp Direct 1stCal w/ Measured 1 stCal w/ Characterized 1sty 8.8 ✓
A
"•A
Frequency (kHz)
Figure 13. Comparison of intermediate ac powers.
108
5.2. VALIDATION ON DESIGN METHODOLOGY
The design methodology is validated using the same experimental prototype, so the
pre-characterization phase has already been conducted. In the initial phase, three different
configurations of the matching capacitance are investigated.
1) Config 1: Ctx = 400 nF and Crx = 200 nF.
2) Config 2: Ctx = 400 nF and Crx = 300 nF.
3) Config 3: Ctx = 300 nF and Crx = 200 nF.
The FOM for each configuration is calculated based on (28) by sweeping the
operating frequency from 120 to 170 kHz, as shown in Figure 14. The global maximum
FOM is achieved with “Config 1” at around 150 kHz. To validate the effectiveness of the
FOM, these configurations are intentionally kept the same in the fine-tune phase. Then, the
other design parameters are tuned based on the initial values set by the local maximum
FOM of each configuration. Figure 15 compares the system efficiencies from both
calculations and experiments. The trend of system efficiency is very close to the predictions
based on the FOM, which implies that the FOM can effectively determine the proper initial
values for the design parameters. Note that the system efficiency is also affected by other
blocks. Therefore, it is necessary to tune the design parameters in fine steps for the optimal
system efficiency. The actual calculations in the fine-tune phase is from 135 to 165 kHz
for all configurations, which is illustrated by the dashed line box in Figure 15. The final
optimal configurations are: Ctx = 400 nF, Crx = 200 nF, Vin = 8.52 V, and operating
frequency at 150 kHz.
109
0.905 Max FOM in Initial Phase
0.895
0.89
0.885 Contig 1Conlig 2Config 3
0.88
Frequency (kHz)
Figure 14. FOM calculation in the initial phase.
Fine-tune phase ^82.5
81/ A
Cal ContigCal Config✓ Cal ConligExp Contig80Exp ConligExp Contig
Frequency (kHz)
Figure 15. System efficiency comparison in the fine-tune phase.
110
In this article, the impractical assumptions of the existing WPT system modeling
methods are investigated by circuit simulations. To avoid the pure resistance load
assumption, an accurate rectifier characterization method is proposed to provide a realistic
load condition. On the basis of the characterized rectifier impedance, a novel system-level
WPT model is presented instead of only focusing on the coil-to-coil part. The improved
model considers both the fundamental frequency and higher order harmonics, which can
effectively eliminate the inaccuracy caused by FHA at off-resonance conditions. Good
performance of the model is validated by experimental measurements in both time and
frequency domains. The result shows that it can accurately estimate the system efficiency
and power loss at each interior block. With the assistance of this model, a practical
methodology is introduced to efficiently optimize the design parameters of a WPT system.
6. CONCLUSION
APPENDIX
Based on the small-signal analysis, the derivations of (6) is described in the
following procedures. All variables are defined the same as Figure 4.
1) Define the source voltage vs with a perturbation: Assume that the source voltage
vs consists of a sinusoidal wave v 1 and a small-signal perturbation Avp : vs = v 1 + Avp,
where jv1j = Vj_, |Avp | = Vp with Vp << Vj_.
111
2) Calculate the output voltage vdc: Using the nonlinear switching functions S and
AS defined in Section 3, the mapping relationship between the input and output sides of the
rectifier can be expressed as
(1)
The input voltage vac and output voltage vdc are calculated by combining the source
voltage vs, the switching functions (S + AS), and the line impedance Zline as
Vac — Va lac * Z\ine= (t'l + A fp) — (S' + AS") • zdc * Z\\ne
Vdc — (S + A S ) ■ vac= (S + A S ) * [(u! + A Up) - (5 + A S ) ■ idC ■ Z llIK ~ S ■ V{ + A S ■ v\ + S • Aup—
S~ ' idc ‘ Z\ine ‘IS ■ A S ■ Z d c ' Z\\nc
(2)
(3)
Note that two terms (AS • Avp) and (AS2 • idc • Zline) are neglected because they involve
the product of two small variation values AS and Avp.
The first term (S • vx) is caused by the main wave. The second and third terms,
(AS • vx) and (S • Avp), are related to the perturbation. The other terms are affected by the
mutual effects of both signals, because the output current idc contains both main wave and
perturbation components. In order to decouple the mutual effects, idc can be approximated
to the dc component of the load current Id. Then, the output voltage vdc can be expressed
by the main wave component vdc, 1 and the perturbation component Avdc
Uic = vdc,i + A v dc, (4)
112
where
(5)
3) Calculate the output current idc: Similar with vdc, the output current idc consists
of the main wave component idc1 and the perturbation component Aidc
?dc — ^'dc,l T~ ^ ^ d c * (6)
The dc component of idc is denoted as Id, which can be simply determined by the rated
output power Pout and the load impedance Rload: Id = Pout /R load-
The ac components are calculated based on the Thevenin equivalent circuit
representing the output side of the rectifier, as shown in Figure A.1. The voltage source of
this equivalent circuit can be either vdc1 or Avdc. Correspondingly, the current is idc1 or
Aidc. The source impedance Zdc is mapped from the ac-side line impedance Zline. For the
single-phase rectifier without overlap period, Zdc approximately equals 2Zline [24], [25].
Then, the ac components of the output current can be calculated as
Figure A.1. Thevenin equivalent circuit for the output side of the rectifier.
113
(7)
(8)
4) Map the output current to the ac-side input current iac: Based on the mapping
relationship, the ac-side input current iac can be calculated from the output current as
iac = (S + AS) ■ idc = (S + A S) ■ (idc.i + A ^c) (9)
To calculate the input impedance of the rectifier, only the perturbation component Ai ac is
needed. So, Aiac is extracted from (37) and expressed as
A iac — A S ' idc.l ~~i~ S ' Aide2SASv\ + S2 Avp - 3S2A S IdZ]me (10)
‘2Z\\ne I i?ioad
The perturbation component of the input voltage Avac is:
A vac ~ A'Vp A S ■ Id ■ ^Une- (11)
Finally, the input impedance of the rectifier can be calculated by the ratio of perturbation
voltage and current at the input side Avac/A iac, as shown in (6).
REFERENCES
[1] C. C. Mi, G. Buja, S. Y. Choi, and C. T. Rim, “Modern advances in wireless power transfer systems for roadway powered electric vehicles,” IEEE Trans. Ind. Electron., vol. 63, no. 10, pp. 6533-6545, Oct. 2016.
[2] S. Li and C. C. Mi, “Wireless power transfer for electric vehicle applications,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 3, no. 1, pp. 4-17, Mar. 2015.
114
[3] X. Lu, P. Wang, D. Niyato, D. I. Kim, and Z. Han, “Wireless charging technologies: Fundamentals, standards, and network applications,” IEEE Commun. Surv. Tut., vol. 18, no. 2, pp. 1413-1452, Apr.-Jun. 2016.
[4] X. Mou and H. Sun, “Wireless power transfer: Survey and roadmap,” in Proc. 81st IEEE Veh. Technol. Conf., 2015, pp. 1-5.
[5] D. Van Wageningen and T. Staring, “The QI wireless power standard,” in Proc. 14th Int. IEEE Power Electron. Motion Control Conf., 2010, pp. S15-25.
[6] J. Schneider, SAE J2954 Overview and Path Forward. Warrendale, PA, USA: SAE Int., 2013.
[7] J. Cho, J. Sun, H. Kim, J. Fan, Y. Lu, and S. Pan, “Coil design for 100 kHz and 6.78 MHz WPT system: Litz and solid wires and winding methods,” in Proc. IEEE Int. Symp. Electromagn. Compat. Signal/Power Integrity, 2017, pp. 803-806.
[8] C. -S. Wang, G. A. Covic, and O. H. Stielau, “Power transfer capability and bifurcation phenomena of loosely coupled inductive power transfer systems,” IEEE Trans. Ind. Electron., vol. 51, no. 1, pp. 148-157, Feb. 2004.
[9] Y. H. Sohn, B. H. Choi, E. S. Lee, G. C. Lim, G.-H. Cho, and C. T. Rim, “General unified analyses of two-capacitor inductive power transfer systems: Equivalence of current-source SS and SP compensations,” IEEE Trans. Power Electron., vol. 30, no. 11, pp. 6030-6045, Nov. 2015.
[10] M. Fu, Z. Tang, and C. Ma, “Analysis and optimized design of compensation capacitors for a megahertz WPT system using full-bridge rectifier,” IEEE Trans. Ind. Inform., vol. 15, no. 1, pp. 95-104, Jan. 2019.
[11] Y. Guo, L. Wang, Y. Zhang, S. Li, and C. Liao, “Rectifier load analysis for electric vehicle wireless charging system,” IEEE Trans. Ind. Electron., vol. 65, no. 9, pp. 6970-6982, Sep. 2018.
[12] Y. Zhang, T. Kan, Z. Yan, and C. C. Mi, “Frequency and voltage tuning of series- series compensated wireless power transfer system to sustain rated power under various conditions,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 7, no. 2, pp. 1311-1317, Jun. 2019.
[13] C. Oeder, A. Bucher, J. Stahl, and T. Duerbaum, “A comparison of different design methods for the multiresonant LLC converter with capacitive output filter,” in Proc. IEEE 12th Workshop Control Model. Power Electron., 2010, pp. 1-7.
[14] M. K. Kazimierczuk and D. Czarkowski, Resonant Power Converters. Hoboken, NJ, USA: Wiley, 2012.
115
[15] L.Jianyu, T. Houjun, and G. Xin, “Frequency splitting analysis of wireless power transfer system based on t-type transformer model,” Elektronika ir Elektrotechnika, vol. 19, no. 10, pp. 109-113, 2013.
[16] T. Kan, T.-D. Nguyen, J. C. White, R. K. Malhan, and C. C. Mi, “A new integration method for an electric vehicle wireless charging system using LCC compensation topology: Analysis and design,” IEEE Trans. power Electron., vol. 32, no. 2, pp. 1638-1650, Feb. 2017.
[17] S. Y. Choi, J. Huh, W. Y. Lee, and C. T. Rim, “Asymmetric coil sets for wireless stationary EV chargers with large lateral tolerance by dominant field analysis,” IEEE Trans. Power Electron., vol. 29, no. 12, pp. 6406-6420, Dec. 2014.
[18] Y. Zhang, K. Chen, F. He, Z. Zhao, T. Lu, and L. Yuan, “Closed-form oriented modeling and analysis of wireless power transfer system with constant-voltage source and load,” IEEE Trans. Power Electron., vol. 31, no. 5, pp. 3472-3481, May 2016.
[19] P. Grzejszczak and R. Barlik, “The description of turn-off process and evaluation of switching power losses in the ultra fast power MOSFET,” Power Electron. Drives, vol. 1, no. 1, pp. 55-67, 2016.
[20] P. Smeets, “First harmonic approximation—Power transfer deviation for resonant LLC converters,” ZeoN PowerTec Tech. Note, 2017.
[21] S. Ladan and K. Wu, “Nonlinear modeling and harmonic recycling of millimeter- wave rectifier circuit,” IEEE Trans. Microw. Theory Techn., vol. 63, no. 3, pp. 937-944, Mar. 2015.
[22] J. Sun and K. J. Karimi, “Small-signal input impedance modeling of line-frequency rectifiers,” IEEE Trans. Aerosp. Electron. Syst., vol. 44, no. 4, pp. 1489-1497, Oct. 2008.
[23] J. Sun, “Small-signal methods for ac distributed power systems—A review,” IEEE Trans. Power Electron., vol. 24, no. 11, pp. 2545-2554, Nov. 2009.
[24] Q. Lei, M. Shen, V. Blasko, and F. Z. Peng, “A generalized input impedance model of three phase diode rectifier,” in Proc. 28th Annu. IEEE Appl. Power Electron. Conf. Expo., 2013, pp. 2655-2661.
[25] M. Sakui, H. Fujita, and M. Shioya, “A method for calculating harmonic currents of a three-phase bridge uncontrolled rectifier with dc filter,” IEEE Trans. Ind. Electron., vol. 36, no. 3, pp. 434-440, Aug. 1989.
[26] T. Kuphaldt, Lessons in Electric Circuits— Semiconductors, vol. III, 2009.
116
[27] W. Zhong and S. Hui, “Maximum energy efficiency tracking for wireless power transfer systems,” IEEE Trans. Power Electron., vol. 30, no. 7, pp. 4025-4034, Jul. 2015.
[28] Y. Yang, W. Zhong, S. Kiratipongvoot, S.-C. Tan, and S. Y. R. Hui, “Dynamic improvement of series-series compensated wireless power transfer systems using discrete sliding mode control,” IEEE Trans. Power Electron., vol. 33, no. 7, pp. 6351-6360, Jul. 2018.
[29] W. Zhong and S. R. Hui, “Charging time control of wireless power transfer systems without using mutual coupling information and wireless communication system,” IEEE Trans. Ind. Electron., vol. 64, no. 1, pp. 228-235, Jan. 2017.
[30] ROHM Semiconductor, “Switching regulator IC series: Calculation of power loss (synchronous),” 2016. Accessed: Apr. 17, 2019. [Online]. Available: http://rohmfs. rohm .com/en/products/databook/applinote/i c/power/switching_regul ator/power_ loss_appli-e.pdf
117
SECTION
2. CONCLUSIONS
With the rapid developments of power electronic devices and technologies, there
are serious design challenges in the modern electrical systems. The modeling techniques
introduced in this dissertation provide powerful tools for the system designers. With the
assistant of the measurement-based characterizations, analytical system-level models can
be established to accurately predict the system performance for both the PDN and WPT
systems. Based on the system-level models, practical methodologies are also proposed to
accelerate the development process at the pre and post design stages.
In the first paper, a novel pattern-based analytical method is proposed for the
impedance calculation. By utilizing the relative relationships between the adjacent vias, the
localized via patterns are formulated based on the physical geometry. Then, the parasitic
via inductance and resistance are analytically derived for arbitrary via patterns. In addition,
a practical modeling methodology is developed to model and guide the PDN design. With
the capability of accurate and layout-free impedance estimation, this methodology is
especially useful for the pre-design stage to accelerate the development process. Finally,
the proposed analytical equations and the modeling methodology have been extensively
validated through the measurements and simulations on a real mobile phone PCB. Good
agreements can be observed from both whole structure and region-by-region comparisons.
In the second paper, a generic behavior model based on the current-mode topology
is developed. This behavior model includes all the key components in the voltage control
118
loop, the current control loop, and the power stage of VRM devices. A novel unifying
method is also proposed for the transition between CCM and DCM, which significantly
simplifies the circuit implementations. The model parameters are optimized through the
measurement-based characterizations. Good accuracy of the proposed model can be
validated using both single-phase and 3-phase VRMs. As a result, the voltage droop
waveforms and the PDN impedance can be accurately predicted in the time-domain and
frequency-domain, respectively. Cascading the proposed VRM model with the PCB-level
and package-level PDN models enables a combined PDN analysis, which is much needed
for modern PDN designs.
In the third paper, an accurate rectifier characterization method is proposed to
provide a realistic load condition. On the basis of the characterized rectifier impedance, a
novel system-level WPT model is presented instead of only focusing on the coil-to-coil
part. The improved model considers both the fundamental frequency and higher order
harmonics, which can effectively eliminate the inaccuracy caused by FHA at off-resonance
conditions. Good performance of the model is validated by experimental measurements in
both time and frequency domains. It shows that the system efficiency and power loss at
each interior block can be accurately estimated. With the assistance of this model, a
practical methodology is introduced to efficiently optimize the design parameters of a WPT
system.
119
BIBLIOGRAPHY
[1] W. D. Becker, J. Eckhardt, R. W. Frech, G. A. Katopis, E. Klink, M. F. McAllister, T. G. McNamara, P. Muench, S. R. Richter, and H. Smith, “Modeling, simulation, and measurement of mid-frequency simultaneous switching noise in computer systems,” IEEE Transactions on Components, Packaging, and Manufacturing Technology: Part B, vol. 21, no. 2, pp. 157-163, 1998.
[2] X. Zhou, P.-L. Wong, P. Xu, F. C. Lee, and A. Q. Huang, “Investigation of candidate vrm topologies for future microprocessors,” IEEE Transactions on Power Electronics, vol. 15, no. 6, pp. 1172-1182, 2000.
[3] Mehas, C. J., Kip D. Coonley, and Charles R. Sullivan. "Converter and inductor design for fast-response microprocessor power delivery." 2000 IEEE 31st Annual Power Electronics Specialists Conference. Conference Proceedings (Cat. No. 00CH37018). Vol. 3. IEEE, 2000.
[4] Mu, Zhen. "Power delivery system: Sufficiency, efficiency, and stability." 9th International Symposium on Quality Electronic Design (isqed 2008). IEEE, 2008.
[5] M. Swaminathan, J. Kim, I. Novak, and J. P. Libous, “Power distribution networks for system-on-package: Status and challenges,” IEEE Transactions on Advanced Packaging, vol. 27, no. 2, pp. 286-300, 2004.
[6] S. Sandler, L. S. Picotest, E. Bogatin, and T. LeCroy, “Vrm modeling: A strategy to survive the collision of three worlds,” November 2018. [Online]. Available: “https://www.signalintegrityjournal.com/articles/1017-vrm-modeling-a-strategy- to-survive-the-collision-of-three-worlds”
[7] S. Baek, P. Pun, and A. Agrawal, “Behavioral model of switching dc-dc converter for improving power delivery network design,” in 2012 IEEE 62nd Electronic Components and Technology Conference. IEEE, 2012, pp. 926-929.
[8] R. Ahmadi, D. Paschedag, and M. Ferdowsi, “Closed-loop input and output impedances of dc-dc switching converters operating in voltage and current mode control,” in IECON 2010-36th Annual Conference on IEEE Industrial Electronics Society. IEEE, 2010, pp. 2311-2316.
[9] J. Xu, S. Bai, B. Zhao, K. Nalla, M. Sapozhnikov, J. L. Drewniak, C. Hwang, and J. Fan, “A novel system-level power integrity transient analysis methodology using simplified cpm model, physics-based equivalent circuit pdn model and small signal vrm model,” in 2019 IEEE International Symposium on Electromagnetic Compatibility, Signal & Power Integrity (EMC+ SIPI). IEEE, 2019, pp. 205-210.
120
[10] K. Lee and A. Barber, “Modeling and analysis of multichip module power supply planes,” IEEE Transactions on Components, Packaging, and Manufacturing Technology: Part B, vol. 18, no. 4, pp. 628-639, 1995.
[11] J. Kim, L. Ren, and J. Fan, “Physics-based inductance extraction for via arrays in parallel planes for power distribution network design,” IEEE Transactions on Microwave Theory and Techniques, vol. 58, no. 9, pp. 2434-2447, 2010.
[12] K. Shringarpure, S. Pan, J. Kim, B. Achkir, B. Archambeault, J. Drewniak, and J. Fan, “Formulation and network reduction to a physics based model for analysis of the power distribution network in a production level multi-layered printed circuit board,” IEEE Transactions on Electromagnetic Compatibility, vol. 58, no. 3, pp. 849-858, 2016.
[13] Zhao, Biyao, et al. "Analytical PDN voltage ripple calculation using simplified equivalent circuit model of PCB PDN." 2015 IEEE Symposium on Electromagnetic Compatibility and Signal Integrity. IEEE, 2015.
[14] S. Bai, B. Zhao, J. Cho, A. Ruehli, S. Connor, M. Cocchini, D. Becker, B. Archambeault, and J. Drewniak, “Plane-pair peec modeling for package power layer nets with inductance extraction,” in 2018 IEEE Symposium on Electromagnetic Compatibility, Signal Integrity and Power Integrity (EMC, SI & PI). IEEE, 2018, pp. 1-17.
[15] J. Kim, W. Lee, Y. Shim, J. Shim, K. Kim, J. S. Pak, and J. Kim, “Chip-package hierarchical power distribution network modeling and analysis based on a segmentation method,” IEEE Transactions on advanced packaging, vol. 33, no. 3, pp. 647-659, 2010.
[16] Zhao, Biyao, et al. "Physics-Based Circuit Modeling Methodology for System Power Integrity Analysis and Design." IEEE Transactions on Electromagnetic Compatibility (2019).
[17] I. Erdin and M. Nakhla, “Mixed circuit/electromagnetic analysis of field coupling to high speed interconnects in inhomogenous medium,” in 1999 IEEE International Symposium on Electromagnetic Compatability. Symposium Record (Cat. No. 99CH36261), vol. 1. IEEE, 1999, pp. 446-449.
[18] M. Feliziani and F. Maradei, “Circuit-oriented fem: Solution of circuit-field coupled problems by circuit equations,” IEEE Transactions on Magnetics, vol. 38, no. 2, pp. 965-968, 2002.
[19] E.-X. Liu, X. Wei, Z. Z. Oo, E.-P. Li, and L.-W. Li, “Modeling of advanced multilayered packages with multiple vias and finite ground planes,” in 2007 IEEE Electrical Performance of Electronic Packaging. IEEE, 2007, pp. 275-278.
121
[20] J. Cho, J. Sun, H. Kim, J. Fan, Y. Lu, and S. Pan, “Coil design for 100 khz and 6.78 mhz wpt system: Litz and solid wires and winding methods,” in Electromagnetic Compatibility & Signal/Power Integrity (EMCSI), 2017 IEEE International Symposium on. IEEE, 2017, pp. 803-806.
[21] C.-S. Wang, G. A. Covic, and O. H. Stielau, “Power transfer capability and bifurcation phenomena of loosely coupled inductive power transfer systems,” IEEE transactions on industrial electronics, vol. 51, no. 1, pp. 148-157, 2004.
[22] Y. H. Sohn, B. H. Choi, E. S. Lee, G. C. Lim, G.-H. Cho, and C. T. Rim, “General unified analyses of two-capacitor inductive power transfer systems: Equivalence of current-source ss and sp compensations,” IEEE Trans. Power Electron, vol. 30, no. 11, pp. 6030-6045, 2015.
[23] M. Fu, Z. Tang, and C. Ma, “Analysis and optimized design of compensation capacitors for a megahertz wpt system using full-bridge rectifier,” IEEE Transactions on Industrial Informatics, 2018.
[24] Y. Guo, L. Wang, Y. Zhang, S. Li, and C. Liao, “Rectifier load analysis for electric vehicle wireless charging system,” IEEE Transactions on Industrial Electronics, vol. 65, no. 9, pp. 6970-6982, 2018.
[25] Y. Zhang, T. Kan, Z. Yan, and C. C. Mi, “Frequency and voltage tuning of series- series compensated wireless power transfer system to sustain rated power under various conditions,” IEEE Journal of Emerging and Selected Topics in Power Electronics, 2018.
122
VITA
Jingdong Sun was born in Jinan, Shandong, China. He received the B.S. degree in
Electronic Information and Communications from the Huazhong University of Science and
Technology, Wuhan, China in 2014. He then joined the EMC Laboratory in the Missouri
University of Science and Technology, Rolla, MO, USA. He received the M.S. and Ph.D
degrees of Electrical Engineering from the Missouri University of Science and Technology
in June 2016 and August 2020, respectively.
His research interests included the modeling of power distribution network and
wireless power transfer systems, signal integrity, and electromagnetic interference. He was
the recipient of the IEEE EMC Hardware Design Award in 2015.
top related